id	sid	tid	token	lemma	pos
ap-948	1	1	ap07_4-5.vp	ap07_4-5.vp	NOUN
ap-948	1	2	1	1	NUM
ap-948	1	3	introduction	introduction	NOUN
ap-948	1	4	face	face	NOUN
ap-948	1	5	detection	detection	NOUN
ap-948	1	6	is	be	AUX
ap-948	1	7	one	one	NUM
ap-948	1	8	of	of	ADP
ap-948	1	9	the	the	DET
ap-948	1	10	most	most	ADV
ap-948	1	11	challenging	challenging	ADJ
ap-948	1	12	tasks	task	NOUN
ap-948	1	13	in	in	ADP
ap-948	1	14	object	object	NOUN
ap-948	1	15	recognition	recognition	NOUN
ap-948	1	16	,	,	PUNCT
ap-948	1	17	because	because	SCONJ
ap-948	1	18	of	of	ADP
ap-948	1	19	the	the	DET
ap-948	1	20	high	high	ADJ
ap-948	1	21	variance	variance	NOUN
ap-948	1	22	among	among	ADP
ap-948	1	23	human	human	ADJ
ap-948	1	24	faces	face	NOUN
ap-948	1	25	,	,	PUNCT
ap-948	1	26	including	include	VERB
ap-948	1	27	facial	facial	ADJ
ap-948	1	28	expression	expression	NOUN
ap-948	1	29	.	.	PUNCT
ap-948	2	1	furthermore	furthermore	ADV
ap-948	2	2	,	,	PUNCT
ap-948	2	3	the	the	DET
ap-948	2	4	typical	typical	ADJ
ap-948	2	5	challenges	challenge	NOUN
ap-948	2	6	of	of	ADP
ap-948	2	7	object	object	NOUN
ap-948	2	8	detection	detection	NOUN
ap-948	2	9	,	,	PUNCT
ap-948	2	10	such	such	ADJ
ap-948	2	11	as	as	ADP
ap-948	2	12	variability	variability	NOUN
ap-948	2	13	in	in	ADP
ap-948	2	14	scale	scale	NOUN
ap-948	2	15	,	,	PUNCT
ap-948	2	16	orientation	orientation	NOUN
ap-948	2	17	and	and	CCONJ
ap-948	2	18	pose	pose	VERB
ap-948	2	19	as	as	ADV
ap-948	2	20	well	well	ADV
ap-948	2	21	as	as	ADP
ap-948	2	22	occlusion	occlusion	NOUN
ap-948	2	23	and	and	CCONJ
ap-948	2	24	lighting	lighting	NOUN
ap-948	2	25	conditions	condition	NOUN
ap-948	2	26	have	have	VERB
ap-948	2	27	to	to	PART
ap-948	2	28	be	be	AUX
ap-948	2	29	treated	treat	VERB
ap-948	2	30	by	by	ADP
ap-948	2	31	a	a	DET
ap-948	2	32	face	face	NOUN
ap-948	2	33	detector	detector	NOUN
ap-948	2	34	.	.	PUNCT
ap-948	3	1	in	in	ADP
ap-948	3	2	recent	recent	ADJ
ap-948	3	3	years	year	NOUN
ap-948	3	4	,	,	PUNCT
ap-948	3	5	several	several	ADJ
ap-948	3	6	methods	method	NOUN
ap-948	3	7	have	have	AUX
ap-948	3	8	been	be	AUX
ap-948	3	9	developed	develop	VERB
ap-948	3	10	for	for	ADP
ap-948	3	11	face	face	NOUN
ap-948	3	12	detection	detection	NOUN
ap-948	3	13	that	that	PRON
ap-948	3	14	deal	deal	VERB
ap-948	3	15	with	with	ADP
ap-948	3	16	some	some	PRON
ap-948	3	17	but	but	CCONJ
ap-948	3	18	not	not	PART
ap-948	3	19	all	all	DET
ap-948	3	20	sources	source	NOUN
ap-948	3	21	of	of	ADP
ap-948	3	22	variance	variance	NOUN
ap-948	3	23	.	.	PUNCT
ap-948	4	1	holistic	holistic	ADJ
ap-948	4	2	approaches	approach	NOUN
ap-948	4	3	such	such	ADJ
ap-948	4	4	as	as	ADP
ap-948	4	5	eigenfaces	eigenface	NOUN
ap-948	4	6	[	[	X
ap-948	4	7	12	12	NUM
ap-948	4	8	]	]	PUNCT
ap-948	4	9	(	(	PUNCT
ap-948	4	10	sometimes	sometimes	ADV
ap-948	4	11	called	call	VERB
ap-948	4	12	image	image	NOUN
ap-948	4	13	-	-	PUNCT
ap-948	4	14	based	base	VERB
ap-948	4	15	[	[	X
ap-948	4	16	4	4	NUM
ap-948	4	17	]	]	PUNCT
ap-948	4	18	or	or	CCONJ
ap-948	4	19	appearance	appearance	NOUN
ap-948	4	20	-	-	PUNCT
ap-948	4	21	based	base	VERB
ap-948	4	22	[	[	X
ap-948	4	23	14	14	NUM
ap-948	4	24	]	]	PUNCT
ap-948	4	25	methods	method	NOUN
ap-948	4	26	in	in	ADP
ap-948	4	27	context	context	NOUN
ap-948	4	28	of	of	ADP
ap-948	4	29	detection	detection	NOUN
ap-948	4	30	and	and	CCONJ
ap-948	4	31	recognition)or	recognition)or	NOUN
ap-948	4	32	the	the	DET
ap-948	4	33	boosted	boost	VERB
ap-948	4	34	cascade	cascade	NOUN
ap-948	4	35	of	of	ADP
ap-948	4	36	simple	simple	ADJ
ap-948	4	37	features	feature	NOUN
ap-948	4	38	[	[	X
ap-948	4	39	13	13	NUM
ap-948	4	40	]	]	PUNCT
ap-948	4	41	perform	perform	VERB
ap-948	4	42	object	object	NOUN
ap-948	4	43	detection	detection	NOUN
ap-948	4	44	by	by	ADP
ap-948	4	45	classifying	classify	VERB
ap-948	4	46	image	image	NOUN
ap-948	4	47	regions	region	NOUN
ap-948	4	48	through	through	ADP
ap-948	4	49	a	a	DET
ap-948	4	50	sliding	slide	VERB
ap-948	4	51	window	window	NOUN
ap-948	4	52	[	[	X
ap-948	4	53	15	15	NUM
ap-948	4	54	]	]	PUNCT
ap-948	4	55	.	.	PUNCT
ap-948	5	1	while	while	SCONJ
ap-948	5	2	they	they	PRON
ap-948	5	3	can	can	AUX
ap-948	5	4	be	be	AUX
ap-948	5	5	made	make	VERB
ap-948	5	6	invariant	invariant	ADJ
ap-948	5	7	to	to	PART
ap-948	5	8	scale	scale	VERB
ap-948	5	9	and	and	CCONJ
ap-948	5	10	lighting	lighting	NOUN
ap-948	5	11	,	,	PUNCT
ap-948	5	12	the	the	DET
ap-948	5	13	major	major	ADJ
ap-948	5	14	drawback	drawback	NOUN
ap-948	5	15	of	of	ADP
ap-948	5	16	these	these	DET
ap-948	5	17	techniques	technique	NOUN
ap-948	5	18	is	be	AUX
ap-948	5	19	that	that	SCONJ
ap-948	5	20	they	they	PRON
ap-948	5	21	can	can	AUX
ap-948	5	22	not	not	PART
ap-948	5	23	deal	deal	VERB
ap-948	5	24	with	with	ADP
ap-948	5	25	different	different	ADJ
ap-948	5	26	rotation	rotation	NOUN
ap-948	5	27	or	or	CCONJ
ap-948	5	28	pose	pose	VERB
ap-948	5	29	.	.	PUNCT
ap-948	6	1	in	in	ADP
ap-948	6	2	addition	addition	NOUN
ap-948	6	3	,	,	PUNCT
ap-948	6	4	facial	facial	ADJ
ap-948	6	5	expression	expression	NOUN
ap-948	6	6	and	and	CCONJ
ap-948	6	7	occlusion	occlusion	NOUN
ap-948	6	8	must	must	AUX
ap-948	6	9	be	be	AUX
ap-948	6	10	treated	treat	VERB
ap-948	6	11	as	as	ADP
ap-948	6	12	intra	intra	ADJ
ap-948	6	13	-	-	ADJ
ap-948	6	14	class	class	ADJ
ap-948	6	15	variance	variance	NOUN
ap-948	6	16	,	,	PUNCT
ap-948	6	17	decreasing	decrease	VERB
ap-948	6	18	the	the	DET
ap-948	6	19	performance	performance	NOUN
ap-948	6	20	of	of	ADP
ap-948	6	21	the	the	DET
ap-948	6	22	classifier	classifier	NOUN
ap-948	6	23	/	/	SYM
ap-948	6	24	detector	detector	NOUN
ap-948	6	25	.	.	PUNCT
ap-948	7	1	on	on	ADP
ap-948	7	2	the	the	DET
ap-948	7	3	other	other	ADJ
ap-948	7	4	hand	hand	NOUN
ap-948	7	5	,	,	PUNCT
ap-948	7	6	feature	feature	NOUN
ap-948	7	7	-	-	PUNCT
ap-948	7	8	based	base	VERB
ap-948	7	9	approaches	approach	NOUN
ap-948	7	10	handle	handle	VERB
ap-948	7	11	pose	pose	NOUN
ap-948	7	12	,	,	PUNCT
ap-948	7	13	expression	expression	NOUN
ap-948	7	14	and	and	CCONJ
ap-948	7	15	even	even	ADV
ap-948	7	16	partial	partial	ADJ
ap-948	7	17	occlusion	occlusion	NOUN
ap-948	7	18	very	very	ADV
ap-948	7	19	well	well	ADV
ap-948	7	20	.	.	PUNCT
ap-948	8	1	rotation	rotation	NOUN
ap-948	8	2	invariance	invariance	NOUN
ap-948	8	3	is	be	AUX
ap-948	8	4	only	only	ADV
ap-948	8	5	limited	limit	VERB
ap-948	8	6	by	by	ADP
ap-948	8	7	the	the	DET
ap-948	8	8	properties	property	NOUN
ap-948	8	9	of	of	ADP
ap-948	8	10	the	the	DET
ap-948	8	11	features	feature	NOUN
ap-948	8	12	used	use	VERB
ap-948	8	13	to	to	PART
ap-948	8	14	build	build	VERB
ap-948	8	15	the	the	DET
ap-948	8	16	detector	detector	NOUN
ap-948	8	17	.	.	PUNCT
ap-948	9	1	this	this	DET
ap-948	9	2	paper	paper	NOUN
ap-948	9	3	presents	present	VERB
ap-948	9	4	a	a	DET
ap-948	9	5	new	new	ADJ
ap-948	9	6	approach	approach	NOUN
ap-948	9	7	to	to	ADP
ap-948	9	8	the	the	DET
ap-948	9	9	detection	detection	NOUN
ap-948	9	10	of	of	ADP
ap-948	9	11	facial	facial	ADJ
ap-948	9	12	features	feature	NOUN
ap-948	9	13	.	.	PUNCT
ap-948	10	1	the	the	DET
ap-948	10	2	local	local	ADJ
ap-948	10	3	features	feature	NOUN
ap-948	10	4	used	use	VERB
ap-948	10	5	for	for	ADP
ap-948	10	6	detection	detection	NOUN
ap-948	10	7	are	be	AUX
ap-948	10	8	invariant	invariant	ADJ
ap-948	10	9	to	to	PART
ap-948	10	10	scale	scale	VERB
ap-948	10	11	,	,	PUNCT
ap-948	10	12	rotation	rotation	NOUN
ap-948	10	13	and	and	CCONJ
ap-948	10	14	change	change	NOUN
ap-948	10	15	in	in	ADP
ap-948	10	16	illumination	illumination	NOUN
ap-948	10	17	,	,	PUNCT
ap-948	10	18	and	and	CCONJ
ap-948	10	19	are	be	AUX
ap-948	10	20	robust	robust	ADJ
ap-948	10	21	to	to	ADP
ap-948	10	22	changing	change	VERB
ap-948	10	23	viewpoints	viewpoint	NOUN
ap-948	10	24	.	.	PUNCT
ap-948	11	1	we	we	PRON
ap-948	11	2	show	show	VERB
ap-948	11	3	that	that	SCONJ
ap-948	11	4	the	the	DET
ap-948	11	5	features	feature	NOUN
ap-948	11	6	chosen	choose	VERB
ap-948	11	7	are	be	AUX
ap-948	11	8	well	well	ADV
ap-948	11	9	suited	suited	ADJ
ap-948	11	10	to	to	PART
ap-948	11	11	detect	detect	VERB
ap-948	11	12	several	several	ADJ
ap-948	11	13	meaningful	meaningful	ADJ
ap-948	11	14	parts	part	NOUN
ap-948	11	15	of	of	ADP
ap-948	11	16	the	the	DET
ap-948	11	17	human	human	ADJ
ap-948	11	18	face	face	NOUN
ap-948	11	19	like	like	ADP
ap-948	11	20	pupils	pupil	NOUN
ap-948	11	21	,	,	PUNCT
ap-948	11	22	nostrils	nostril	NOUN
ap-948	11	23	and	and	CCONJ
ap-948	11	24	the	the	DET
ap-948	11	25	corner	corner	NOUN
ap-948	11	26	of	of	ADP
ap-948	11	27	the	the	DET
ap-948	11	28	mouth	mouth	NOUN
ap-948	11	29	.	.	PUNCT
ap-948	12	1	the	the	DET
ap-948	12	2	paper	paper	NOUN
ap-948	12	3	is	be	AUX
ap-948	12	4	organized	organize	VERB
ap-948	12	5	as	as	SCONJ
ap-948	12	6	follows	follow	VERB
ap-948	12	7	:	:	PUNCT
ap-948	12	8	in	in	ADP
ap-948	12	9	section	section	NOUN
ap-948	12	10	two	two	NUM
ap-948	12	11	,	,	PUNCT
ap-948	12	12	the	the	DET
ap-948	12	13	feature	feature	NOUN
ap-948	12	14	extraction	extraction	NOUN
ap-948	12	15	process	process	NOUN
ap-948	12	16	is	be	AUX
ap-948	12	17	described	describe	VERB
ap-948	12	18	.	.	PUNCT
ap-948	13	1	feature	feature	NOUN
ap-948	13	2	space	space	NOUN
ap-948	13	3	reduction	reduction	NOUN
ap-948	13	4	and	and	CCONJ
ap-948	13	5	feature	feature	NOUN
ap-948	13	6	selection	selection	NOUN
ap-948	13	7	are	be	AUX
ap-948	13	8	explained	explain	VERB
ap-948	13	9	in	in	ADP
ap-948	13	10	section	section	NOUN
ap-948	13	11	three	three	NUM
ap-948	13	12	.	.	PUNCT
ap-948	14	1	section	section	NOUN
ap-948	14	2	four	four	NUM
ap-948	14	3	presents	present	VERB
ap-948	14	4	the	the	DET
ap-948	14	5	classifier	classifier	NOUN
ap-948	14	6	,	,	PUNCT
ap-948	14	7	and	and	CCONJ
ap-948	14	8	the	the	DET
ap-948	14	9	results	result	NOUN
ap-948	14	10	are	be	AUX
ap-948	14	11	shown	show	VERB
ap-948	14	12	in	in	ADP
ap-948	14	13	section	section	NOUN
ap-948	14	14	five	five	NUM
ap-948	14	15	.	.	PUNCT
ap-948	15	1	in	in	ADP
ap-948	15	2	section	section	NOUN
ap-948	15	3	six	six	NUM
ap-948	15	4	,	,	PUNCT
ap-948	15	5	the	the	DET
ap-948	15	6	results	result	NOUN
ap-948	15	7	are	be	AUX
ap-948	15	8	discussed	discuss	VERB
ap-948	15	9	.	.	PUNCT
ap-948	16	1	2	2	NUM
ap-948	16	2	feature	feature	NOUN
ap-948	16	3	extraction	extraction	NOUN
ap-948	16	4	this	this	DET
ap-948	16	5	section	section	NOUN
ap-948	16	6	explains	explain	VERB
ap-948	16	7	feature	feature	NOUN
ap-948	16	8	extraction	extraction	NOUN
ap-948	16	9	from	from	ADP
ap-948	16	10	an	an	DET
ap-948	16	11	image	image	NOUN
ap-948	16	12	and	and	CCONJ
ap-948	16	13	its	its	PRON
ap-948	16	14	description	description	NOUN
ap-948	16	15	in	in	ADP
ap-948	16	16	feature	feature	NOUN
ap-948	16	17	space	space	NOUN
ap-948	16	18	,	,	PUNCT
ap-948	16	19	the	the	DET
ap-948	16	20	features	feature	NOUN
ap-948	16	21	have	have	VERB
ap-948	16	22	to	to	PART
ap-948	16	23	be	be	AUX
ap-948	16	24	invariant	invariant	ADJ
ap-948	16	25	to	to	PART
ap-948	16	26	affine	affine	VERB
ap-948	16	27	transformations	transformation	NOUN
ap-948	16	28	and	and	CCONJ
ap-948	16	29	illumination	illumination	NOUN
ap-948	16	30	.	.	PUNCT
ap-948	17	1	while	while	SCONJ
ap-948	17	2	there	there	PRON
ap-948	17	3	have	have	AUX
ap-948	17	4	been	be	AUX
ap-948	17	5	numerous	numerous	ADJ
ap-948	17	6	proposals	proposal	NOUN
ap-948	17	7	for	for	ADP
ap-948	17	8	appropriate	appropriate	ADJ
ap-948	17	9	feature	feature	NOUN
ap-948	17	10	points	point	NOUN
ap-948	17	11	,	,	PUNCT
ap-948	17	12	they	they	PRON
ap-948	17	13	have	have	AUX
ap-948	17	14	usually	usually	ADV
ap-948	17	15	been	be	AUX
ap-948	17	16	chosen	choose	VERB
ap-948	17	17	on	on	ADP
ap-948	17	18	the	the	DET
ap-948	17	19	basis	basis	NOUN
ap-948	17	20	of	of	ADP
ap-948	17	21	human	human	ADJ
ap-948	17	22	observation	observation	NOUN
ap-948	17	23	of	of	ADP
ap-948	17	24	face	face	NOUN
ap-948	17	25	characteristics	characteristic	NOUN
ap-948	17	26	.	.	PUNCT
ap-948	18	1	in	in	ADP
ap-948	18	2	contrast	contrast	NOUN
ap-948	18	3	to	to	ADP
ap-948	18	4	such	such	ADJ
ap-948	18	5	knowledge	knowledge	NOUN
ap-948	18	6	-	-	PUNCT
ap-948	18	7	based	base	VERB
ap-948	18	8	methods	method	NOUN
ap-948	18	9	,	,	PUNCT
ap-948	18	10	several	several	ADJ
ap-948	18	11	methods	method	NOUN
ap-948	18	12	have	have	AUX
ap-948	18	13	been	be	AUX
ap-948	18	14	developed	develop	VERB
ap-948	18	15	to	to	PART
ap-948	18	16	detect	detect	VERB
ap-948	18	17	structures	structure	NOUN
ap-948	18	18	that	that	PRON
ap-948	18	19	are	be	AUX
ap-948	18	20	generally	generally	ADV
ap-948	18	21	easy	easy	ADJ
ap-948	18	22	to	to	PART
ap-948	18	23	locate	locate	VERB
ap-948	18	24	,	,	PUNCT
ap-948	18	25	can	can	AUX
ap-948	18	26	be	be	AUX
ap-948	18	27	computed	compute	VERB
ap-948	18	28	with	with	ADP
ap-948	18	29	high	high	ADJ
ap-948	18	30	reliability	reliability	NOUN
ap-948	18	31	and	and	CCONJ
ap-948	18	32	satisfy	satisfy	VERB
ap-948	18	33	the	the	DET
ap-948	18	34	demand	demand	NOUN
ap-948	18	35	of	of	ADP
ap-948	18	36	scale	scale	NOUN
ap-948	18	37	,	,	PUNCT
ap-948	18	38	rotation	rotation	NOUN
ap-948	18	39	and	and	CCONJ
ap-948	18	40	illumination	illumination	NOUN
ap-948	18	41	invariance	invariance	NOUN
ap-948	19	1	[	[	X
ap-948	19	2	6	6	NUM
ap-948	19	3	,	,	PUNCT
ap-948	19	4	10	10	NUM
ap-948	19	5	]	]	PUNCT
ap-948	19	6	.	.	PUNCT
ap-948	20	1	the	the	DET
ap-948	20	2	local	local	ADJ
ap-948	20	3	area	area	NOUN
ap-948	20	4	around	around	ADP
ap-948	20	5	these	these	DET
ap-948	20	6	so	so	ADV
ap-948	20	7	-	-	PUNCT
ap-948	20	8	called	call	VERB
ap-948	20	9	interest	interest	NOUN
ap-948	20	10	points	point	NOUN
ap-948	20	11	is	be	AUX
ap-948	20	12	then	then	ADV
ap-948	20	13	extracted	extract	VERB
ap-948	20	14	and	and	CCONJ
ap-948	20	15	modeled	model	VERB
ap-948	20	16	.	.	PUNCT
ap-948	21	1	a	a	DET
ap-948	21	2	3d	3d	NUM
ap-948	21	3	scale	scale	NOUN
ap-948	21	4	-	-	PUNCT
ap-948	21	5	space	space	NOUN
ap-948	21	6	representation	representation	NOUN
ap-948	21	7	of	of	ADP
ap-948	21	8	an	an	DET
ap-948	21	9	image	image	NOUN
ap-948	21	10	is	be	AUX
ap-948	21	11	usually	usually	ADV
ap-948	21	12	taken	take	VERB
ap-948	21	13	to	to	PART
ap-948	21	14	detect	detect	VERB
ap-948	21	15	local	local	ADJ
ap-948	21	16	features	feature	NOUN
ap-948	21	17	and	and	CCONJ
ap-948	21	18	their	their	PRON
ap-948	21	19	corresponding	correspond	VERB
ap-948	21	20	scales	scale	NOUN
ap-948	21	21	(	(	PUNCT
ap-948	21	22	see	see	VERB
ap-948	21	23	fig	fig	NOUN
ap-948	21	24	.	.	PUNCT
ap-948	22	1	1	1	NUM
ap-948	22	2	)	)	PUNCT
ap-948	22	3	.	.	PUNCT
ap-948	23	1	given	give	VERB
ap-948	23	2	any	any	DET
ap-948	23	3	image	image	NOUN
ap-948	23	4	i(x	i(x	NOUN
ap-948	23	5	)	)	PUNCT
ap-948	23	6	,	,	PUNCT
ap-948	23	7	its	its	PRON
ap-948	23	8	scale	scale	NOUN
ap-948	23	9	-	-	PUNCT
ap-948	23	10	space	space	NOUN
ap-948	23	11	representation	representation	NOUN
ap-948	23	12	l(x	l(x	PROPN
ap-948	23	13	,	,	PUNCT
ap-948	23	14	�	�	PROPN
ap-948	23	15	)	)	PUNCT
ap-948	23	16	,	,	PUNCT
ap-948	23	17	is	be	AUX
ap-948	23	18	defined	define	VERB
ap-948	23	19	by	by	ADP
ap-948	23	20	l	l	PROPN
ap-948	23	21	g	g	PROPN
ap-948	23	22	i	i	PROPN
ap-948	23	23	(	(	PUNCT
ap-948	23	24	,	,	PUNCT
ap-948	23	25	)	)	PUNCT
ap-948	23	26	(	(	PUNCT
ap-948	23	27	,	,	PUNCT
ap-948	23	28	)	)	PUNCT
ap-948	23	29	(	(	PUNCT
ap-948	23	30	)	)	PUNCT
ap-948	23	31	x	x	SYM
ap-948	23	32	x	x	SYM
ap-948	23	33	x	x	X
ap-948	23	34	�	�	PROPN
ap-948	23	35	�	�	PROPN
ap-948	23	36	�	�	PROPN
ap-948	23	37	�	�	PROPN
ap-948	23	38	(	(	PUNCT
ap-948	23	39	1	1	NUM
ap-948	23	40	)	)	PUNCT
ap-948	23	41	where	where	SCONJ
ap-948	23	42	g	g	NOUN
ap-948	23	43	(	(	PUNCT
ap-948	23	44	,	,	PUNCT
ap-948	23	45	)	)	PUNCT
ap-948	23	46	x	x	PROPN
ap-948	23	47	�	�	PROPN
ap-948	23	48	denotes	denote	VERB
ap-948	23	49	the	the	DET
ap-948	23	50	gaussian	gaussian	ADJ
ap-948	23	51	kernel	kernel	NOUN
ap-948	23	52	function	function	NOUN
ap-948	23	53	given	give	VERB
ap-948	23	54	by	by	ADP
ap-948	23	55	g	g	PROPN
ap-948	23	56	(	(	PUNCT
ap-948	23	57	,	,	PUNCT
ap-948	23	58	)	)	PUNCT
ap-948	23	59	expx	expx	PROPN
ap-948	23	60	x	x	SYM
ap-948	23	61	�	�	PROPN
ap-948	23	62	�	�	PROPN
ap-948	23	63	�	�	PROPN
ap-948	23	64	�	�	PROPN
ap-948	23	65	�	�	PROPN
ap-948	23	66	�	�	PROPN
ap-948	23	67	�	�	PROPN
ap-948	23	68	�	�	PROPN
ap-948	23	69	�	�	PROPN
ap-948	23	70	�	�	PROPN
ap-948	23	71	�	�	PROPN
ap-948	23	72	�	�	PROPN
ap-948	23	73	1	1	NUM
ap-948	23	74	2	2	NUM
ap-948	23	75	2	2	NUM
ap-948	23	76	2	2	NUM
ap-948	23	77	2	2	NUM
ap-948	23	78	2	2	NUM
ap-948	23	79	.	.	PUNCT
ap-948	24	1	(	(	PUNCT
ap-948	24	2	2	2	X
ap-948	24	3	)	)	PUNCT
ap-948	24	4	our	our	PRON
ap-948	24	5	approach	approach	NOUN
ap-948	24	6	uses	use	VERB
ap-948	24	7	the	the	DET
ap-948	24	8	scale	scale	NOUN
ap-948	24	9	adapted	adapt	VERB
ap-948	24	10	harris	harris	PROPN
ap-948	24	11	corner	corner	PROPN
ap-948	24	12	detector	detector	NOUN
ap-948	24	13	to	to	PART
ap-948	24	14	detect	detect	VERB
ap-948	24	15	interest	interest	NOUN
ap-948	24	16	points	point	NOUN
ap-948	24	17	.	.	PUNCT
ap-948	25	1	this	this	DET
ap-948	25	2	detector	detector	NOUN
ap-948	25	3	is	be	AUX
ap-948	25	4	based	base	VERB
ap-948	25	5	on	on	ADP
ap-948	25	6	the	the	DET
ap-948	25	7	second	second	ADJ
ap-948	25	8	moment	moment	NOUN
ap-948	25	9	matrix	matrix	NOUN
ap-948	25	10	[	[	X
ap-948	25	11	7	7	NUM
ap-948	25	12	]	]	X
ap-948	25	13	:	:	PUNCT
ap-948	25	14	�	�	PROPN
ap-948	25	15	(	(	PUNCT
ap-948	25	16	,	,	PUNCT
ap-948	25	17	,	,	PUNCT
ap-948	25	18	)	)	PUNCT
ap-948	25	19	(	(	PUNCT
ap-948	25	20	)	)	PUNCT
ap-948	25	21	(	(	PUNCT
ap-948	25	22	,	,	PUNCT
ap-948	25	23	)	)	PUNCT
ap-948	25	24	(	(	PUNCT
ap-948	25	25	,	,	PUNCT
ap-948	25	26	)	)	PUNCT
ap-948	25	27	(	(	PUNCT
ap-948	25	28	,	,	PUNCT
ap-948	25	29	)	)	PUNCT
ap-948	25	30	x	x	PUNCT
ap-948	26	1	x	x	PUNCT
ap-948	26	2	x	x	SYM
ap-948	26	3	x	x	SYM
ap-948	26	4	�	�	PROPN
ap-948	26	5	�	�	PROPN
ap-948	26	6	�	�	PROPN
ap-948	26	7	�	�	PROPN
ap-948	26	8	�	�	PROPN
ap-948	26	9	�	�	PROPN
ap-948	26	10	�	�	PROPN
ap-948	27	1	i	i	NOUN
ap-948	27	2	d	d	PROPN
ap-948	28	1	d	d	NOUN
ap-948	28	2	i	i	NOUN
ap-948	29	1	d	d	NOUN
ap-948	29	2	d	d	X
ap-948	29	3	x	x	X
ap-948	29	4	y	y	NOUN
ap-948	29	5	d	d	NOUN
ap-948	29	6	x	x	X
ap-948	29	7	y	y	NOUN
ap-948	30	1	d	d	X
ap-948	30	2	g	g	PROPN
ap-948	30	3	l	l	NOUN
ap-948	30	4	l	l	NOUN
ap-948	30	5	l	l	NOUN
ap-948	30	6	l	l	NOUN
ap-948	30	7	l	l	X
ap-948	30	8	l	l	X
ap-948	30	9	�	�	PROPN
ap-948	30	10	�	�	PROPN
ap-948	30	11	2	2	NUM
ap-948	30	12	2	2	NUM
ap-948	30	13	d	d	NOUN
ap-948	30	14	d	d	SYM
ap-948	30	15	2	2	NUM
ap-948	30	16	(	(	PUNCT
ap-948	30	17	,	,	PUNCT
ap-948	30	18	)	)	PUNCT
ap-948	30	19	x	x	PROPN
ap-948	30	20	�	�	PROPN
ap-948	30	21	�	�	PROPN
ap-948	30	22	�	�	PROPN
ap-948	30	23	�	�	PROPN
ap-948	30	24	�	�	PROPN
ap-948	30	25	�	�	PROPN
ap-948	30	26	(	(	PUNCT
ap-948	30	27	3	3	NUM
ap-948	30	28	)	)	PUNCT
ap-948	30	29	this	this	DET
ap-948	30	30	matrix	matrix	NOUN
ap-948	30	31	describes	describe	VERB
ap-948	30	32	the	the	DET
ap-948	30	33	gradient	gradient	ADJ
ap-948	30	34	distribution	distribution	NOUN
ap-948	30	35	in	in	ADP
ap-948	30	36	a	a	DET
ap-948	30	37	local	local	ADJ
ap-948	30	38	neighborhood	neighborhood	NOUN
ap-948	30	39	of	of	ADP
ap-948	30	40	a	a	DET
ap-948	30	41	point	point	NOUN
ap-948	30	42	.	.	PUNCT
ap-948	31	1	its	its	PRON
ap-948	31	2	eigenvalues	eigenvalue	NOUN
ap-948	31	3	�	�	PROPN
ap-948	31	4	1	1	NUM
ap-948	31	5	,	,	PUNCT
ap-948	31	6	�	�	PROPN
ap-948	31	7	2	2	NUM
ap-948	31	8	represent	represent	VERB
ap-948	31	9	the	the	DET
ap-948	31	10	two	two	NUM
ap-948	31	11	principal	principal	ADJ
ap-948	31	12	signal	signal	NOUN
ap-948	31	13	changes	change	NOUN
ap-948	31	14	.	.	PUNCT
ap-948	32	1	thus	thus	ADV
ap-948	32	2	it	it	PRON
ap-948	32	3	is	be	AUX
ap-948	32	4	possible	possible	ADJ
ap-948	32	5	to	to	PART
ap-948	32	6	extract	extract	VERB
ap-948	32	7	points	point	NOUN
ap-948	32	8	with	with	ADP
ap-948	32	9	a	a	DET
ap-948	32	10	significant	significant	ADJ
ap-948	32	11	signal	signal	NOUN
ap-948	32	12	change	change	NOUN
ap-948	32	13	in	in	ADP
ap-948	32	14	both	both	DET
ap-948	32	15	orthogonal	orthogonal	ADJ
ap-948	32	16	directions	direction	NOUN
ap-948	32	17	indicating	indicate	VERB
ap-948	32	18	edges	edge	NOUN
ap-948	32	19	or	or	CCONJ
ap-948	32	20	junctions	junction	NOUN
ap-948	32	21	,	,	PUNCT
ap-948	32	22	for	for	ADP
ap-948	32	23	example	example	NOUN
ap-948	32	24	.	.	PUNCT
ap-948	33	1	since	since	SCONJ
ap-948	33	2	it	it	PRON
ap-948	33	3	©	©	PROPN
ap-948	33	4	czech	czech	PROPN
ap-948	33	5	technical	technical	PROPN
ap-948	33	6	university	university	PROPN
ap-948	33	7	publishing	publishing	NOUN
ap-948	33	8	house	house	NOUN
ap-948	33	9	http://ctn.cvut.cz/ap/	http://ctn.cvut.cz/ap/	PROPN
ap-948	33	10	3	3	NUM
ap-948	33	11	acta	acta	PROPN
ap-948	33	12	polytechnica	polytechnica	PROPN
ap-948	33	13	vol	vol	NOUN
ap-948	33	14	.	.	PUNCT
ap-948	34	1	47	47	NUM
ap-948	34	2	no	no	NOUN
ap-948	34	3	.	.	PUNCT
ap-948	35	1	4–5/2007	4–5/2007	NUM
ap-948	35	2	detection	detection	NOUN
ap-948	35	3	of	of	ADP
ap-948	35	4	facial	facial	ADJ
ap-948	35	5	features	feature	NOUN
ap-948	35	6	in	in	ADP
ap-948	35	7	scale	scale	NOUN
ap-948	35	8	-	-	PUNCT
ap-948	35	9	space	space	NOUN
ap-948	35	10	p.	p.	NOUN
ap-948	35	11	hosten	hosten	NOUN
ap-948	35	12	,	,	PUNCT
ap-948	35	13	m.	m.	NOUN
ap-948	35	14	asbach	asbach	NOUN
ap-948	35	15	this	this	DET
ap-948	35	16	paper	paper	NOUN
ap-948	35	17	presents	present	VERB
ap-948	35	18	a	a	DET
ap-948	35	19	new	new	ADJ
ap-948	35	20	approach	approach	NOUN
ap-948	35	21	to	to	ADP
ap-948	35	22	the	the	DET
ap-948	35	23	detection	detection	NOUN
ap-948	35	24	of	of	ADP
ap-948	35	25	facial	facial	ADJ
ap-948	35	26	features	feature	NOUN
ap-948	35	27	.	.	PUNCT
ap-948	36	1	a	a	DET
ap-948	36	2	scale	scale	NOUN
ap-948	36	3	adapted	adapt	VERB
ap-948	36	4	harris	harris	PROPN
ap-948	36	5	corner	corner	PROPN
ap-948	36	6	detector	detector	NOUN
ap-948	36	7	is	be	AUX
ap-948	36	8	used	use	VERB
ap-948	36	9	to	to	PART
ap-948	36	10	find	find	VERB
ap-948	36	11	interest	interest	NOUN
ap-948	36	12	points	point	NOUN
ap-948	36	13	in	in	ADP
ap-948	36	14	scale	scale	NOUN
ap-948	36	15	-	-	PUNCT
ap-948	36	16	space	space	NOUN
ap-948	36	17	.	.	PUNCT
ap-948	37	1	these	these	DET
ap-948	37	2	points	point	NOUN
ap-948	37	3	are	be	AUX
ap-948	37	4	described	describe	VERB
ap-948	37	5	by	by	ADP
ap-948	37	6	the	the	DET
ap-948	37	7	sift	sift	PROPN
ap-948	37	8	descriptor	descriptor	NOUN
ap-948	37	9	.	.	PUNCT
ap-948	38	1	thus	thus	ADV
ap-948	38	2	invariance	invariance	VERB
ap-948	38	3	with	with	ADP
ap-948	38	4	respect	respect	NOUN
ap-948	38	5	to	to	ADP
ap-948	38	6	image	image	NOUN
ap-948	38	7	scale	scale	NOUN
ap-948	38	8	,	,	PUNCT
ap-948	38	9	rotation	rotation	NOUN
ap-948	38	10	and	and	CCONJ
ap-948	38	11	illumination	illumination	NOUN
ap-948	38	12	is	be	AUX
ap-948	38	13	obtained	obtain	VERB
ap-948	38	14	.	.	PUNCT
ap-948	39	1	applying	apply	VERB
ap-948	39	2	a	a	DET
ap-948	39	3	karhunen	karhunen	NOUN
ap-948	39	4	-	-	PUNCT
ap-948	39	5	loeve	loeve	NOUN
ap-948	39	6	transform	transform	NOUN
ap-948	39	7	reduces	reduce	VERB
ap-948	39	8	the	the	DET
ap-948	39	9	dimensionality	dimensionality	NOUN
ap-948	39	10	of	of	ADP
ap-948	39	11	the	the	DET
ap-948	39	12	feature	feature	NOUN
ap-948	39	13	space	space	NOUN
ap-948	39	14	.	.	PUNCT
ap-948	40	1	in	in	ADP
ap-948	40	2	the	the	DET
ap-948	40	3	training	training	NOUN
ap-948	40	4	process	process	NOUN
ap-948	40	5	these	these	DET
ap-948	40	6	features	feature	NOUN
ap-948	40	7	are	be	AUX
ap-948	40	8	clustered	cluster	VERB
ap-948	40	9	by	by	ADP
ap-948	40	10	the	the	DET
ap-948	40	11	k	k	PROPN
ap-948	40	12	-	-	PUNCT
ap-948	40	13	means	means	NOUN
ap-948	40	14	algorithm	algorithm	NOUN
ap-948	40	15	,	,	PUNCT
ap-948	40	16	followed	follow	VERB
ap-948	40	17	by	by	ADP
ap-948	40	18	a	a	DET
ap-948	40	19	cluster	cluster	NOUN
ap-948	40	20	analysis	analysis	NOUN
ap-948	40	21	to	to	PART
ap-948	40	22	find	find	VERB
ap-948	40	23	the	the	DET
ap-948	40	24	most	most	ADV
ap-948	40	25	distinctive	distinctive	ADJ
ap-948	40	26	clusters	cluster	NOUN
ap-948	40	27	,	,	PUNCT
ap-948	40	28	which	which	PRON
ap-948	40	29	represent	represent	VERB
ap-948	40	30	facial	facial	ADJ
ap-948	40	31	features	feature	NOUN
ap-948	40	32	in	in	ADP
ap-948	40	33	feature	feature	NOUN
ap-948	40	34	space	space	NOUN
ap-948	40	35	.	.	PUNCT
ap-948	41	1	finally	finally	ADV
ap-948	41	2	,	,	PUNCT
ap-948	41	3	a	a	DET
ap-948	41	4	classifier	classifier	NOUN
ap-948	41	5	based	base	VERB
ap-948	41	6	on	on	ADP
ap-948	41	7	the	the	DET
ap-948	41	8	nearest	near	ADJ
ap-948	41	9	neighbor	neighbor	NOUN
ap-948	41	10	approach	approach	NOUN
ap-948	41	11	is	be	AUX
ap-948	41	12	used	use	VERB
ap-948	41	13	to	to	PART
ap-948	41	14	decide	decide	VERB
ap-948	41	15	whether	whether	SCONJ
ap-948	41	16	the	the	DET
ap-948	41	17	features	feature	NOUN
ap-948	41	18	obtained	obtain	VERB
ap-948	41	19	from	from	ADP
ap-948	41	20	the	the	DET
ap-948	41	21	interest	interest	NOUN
ap-948	41	22	points	point	NOUN
ap-948	41	23	are	be	AUX
ap-948	41	24	facial	facial	ADJ
ap-948	41	25	features	feature	NOUN
ap-948	41	26	or	or	CCONJ
ap-948	41	27	not	not	PART
ap-948	41	28	.	.	PUNCT
ap-948	42	1	keywords	keyword	NOUN
ap-948	42	2	:	:	PUNCT
ap-948	42	3	clustering	cluster	VERB
ap-948	42	4	methods	method	NOUN
ap-948	42	5	,	,	PUNCT
ap-948	42	6	face	face	NOUN
ap-948	42	7	recognition	recognition	NOUN
ap-948	42	8	,	,	PUNCT
ap-948	42	9	feature	feature	NOUN
ap-948	42	10	extraction	extraction	NOUN
ap-948	42	11	,	,	PUNCT
ap-948	42	12	interest	interest	NOUN
ap-948	42	13	points	point	NOUN
ap-948	42	14	,	,	PUNCT
ap-948	42	15	karhunen	karhunen	NOUN
ap-948	42	16	-	-	PUNCT
ap-948	42	17	loeve	loeve	NOUN
ap-948	42	18	transforms	transform	VERB
ap-948	42	19	,	,	PUNCT
ap-948	42	20	object	object	VERB
ap-948	42	21	detection	detection	NOUN
ap-948	42	22	,	,	PUNCT
ap-948	42	23	pattern	pattern	NOUN
ap-948	42	24	classification	classification	NOUN
ap-948	42	25	.	.	PUNCT
ap-948	43	1	fig	fig	NOUN
ap-948	43	2	.	.	PUNCT
ap-948	44	1	1	1	NUM
ap-948	44	2	:	:	PUNCT
ap-948	44	3	the	the	DET
ap-948	44	4	image	image	NOUN
ap-948	44	5	i(x	i(x	NOUN
ap-948	44	6	)	)	PUNCT
ap-948	44	7	is	be	AUX
ap-948	44	8	embedded	embed	VERB
ap-948	44	9	into	into	ADP
ap-948	44	10	a	a	DET
ap-948	44	11	continuous	continuous	ADJ
ap-948	44	12	family	family	NOUN
ap-948	44	13	l(x	l(x	PROPN
ap-948	44	14	,	,	PUNCT
ap-948	44	15	�	�	PROPN
ap-948	44	16	)	)	PUNCT
ap-948	44	17	of	of	ADP
ap-948	44	18	gradually	gradually	ADV
ap-948	44	19	smoother	smooth	ADJ
ap-948	44	20	versions	version	NOUN
ap-948	44	21	of	of	ADP
ap-948	44	22	it	it	PRON
ap-948	44	23	.	.	PUNCT
ap-948	45	1	hereby	hereby	ADV
ap-948	45	2	the	the	DET
ap-948	45	3	original	original	ADJ
ap-948	45	4	image	image	NOUN
ap-948	45	5	corresponds	correspond	VERB
ap-948	45	6	to	to	ADP
ap-948	45	7	the	the	DET
ap-948	45	8	scale	scale	NOUN
ap-948	45	9	�	�	PROPN
ap-948	45	10	�	�	PROPN
ap-948	45	11	0	0	PUNCT
ap-948	46	1	[	[	X
ap-948	46	2	11	11	NUM
ap-948	46	3	]	]	PUNCT
ap-948	46	4	.	.	PUNCT
ap-948	47	1	interest	interest	NOUN
ap-948	47	2	points	point	NOUN
ap-948	47	3	are	be	AUX
ap-948	47	4	detected	detect	VERB
ap-948	47	5	in	in	ADP
ap-948	47	6	this	this	DET
ap-948	47	7	scale	scale	NOUN
ap-948	47	8	-	-	PUNCT
ap-948	47	9	space	space	NOUN
ap-948	47	10	.	.	PUNCT
ap-948	48	1	is	be	AUX
ap-948	48	2	easier	easy	ADJ
ap-948	48	3	to	to	PART
ap-948	48	4	compute	compute	VERB
ap-948	48	5	the	the	DET
ap-948	48	6	trace	trace	NOUN
ap-948	48	7	and	and	CCONJ
ap-948	48	8	the	the	DET
ap-948	48	9	determinant	determinant	NOUN
ap-948	48	10	of	of	ADP
ap-948	48	11	the	the	DET
ap-948	48	12	second	second	ADJ
ap-948	48	13	moment	moment	NOUN
ap-948	48	14	matrix	matrix	NOUN
ap-948	48	15	than	than	ADP
ap-948	48	16	its	its	PRON
ap-948	48	17	eigenvalues	eigenvalue	NOUN
ap-948	48	18	,	,	PUNCT
ap-948	48	19	the	the	DET
ap-948	48	20	harris	harris	PROPN
ap-948	48	21	detector	detector	NOUN
ap-948	48	22	uses	use	VERB
ap-948	48	23	the	the	DET
ap-948	48	24	following	follow	VERB
ap-948	48	25	measure	measure	NOUN
ap-948	48	26	to	to	PART
ap-948	48	27	determine	determine	VERB
ap-948	48	28	the	the	DET
ap-948	48	29	location	location	NOUN
ap-948	48	30	of	of	ADP
ap-948	48	31	interest	interest	NOUN
ap-948	48	32	points	point	NOUN
ap-948	49	1	[	[	X
ap-948	49	2	3	3	NUM
ap-948	49	3	]	]	NOUN
ap-948	49	4	:	:	PUNCT
ap-948	49	5	cornerness	cornerness	NOUN
ap-948	49	6	tracei	tracei	PROPN
ap-948	49	7	d	d	PROPN
ap-948	49	8	�	�	PROPN
ap-948	49	9	�	�	PROPN
ap-948	49	10	�	�	PROPN
ap-948	49	11	�	�	PROPN
ap-948	49	12	�	�	PROPN
ap-948	49	13	�	�	PROPN
ap-948	49	14	�	�	PROPN
ap-948	49	15	�	�	PROPN
ap-948	49	16	�	�	PROPN
ap-948	49	17	�	�	PROPN
ap-948	49	18	�	�	PROPN
ap-948	49	19	�	�	PROPN
ap-948	49	20	�	�	PROPN
ap-948	49	21	1	1	NUM
ap-948	49	22	2	2	NUM
ap-948	49	23	1	1	NUM
ap-948	49	24	2	2	NUM
ap-948	49	25	2	2	NUM
ap-948	49	26	2	2	NUM
ap-948	49	27	(	(	PUNCT
ap-948	49	28	)	)	PUNCT
ap-948	49	29	det	det	PROPN
ap-948	49	30	(	(	PUNCT
ap-948	49	31	(	(	PUNCT
ap-948	49	32	,	,	PUNCT
ap-948	49	33	,	,	PUNCT
ap-948	49	34	)	)	PUNCT
ap-948	49	35	)	)	PUNCT
ap-948	50	1	(	(	PUNCT
ap-948	50	2	�	�	PROPN
ap-948	50	3	x	x	SYM
ap-948	50	4	�	�	PROPN
ap-948	50	5	(	(	PUNCT
ap-948	50	6	,	,	PUNCT
ap-948	50	7	,	,	PUNCT
ap-948	50	8	)	)	PUNCT
ap-948	50	9	)	)	PUNCT
ap-948	50	10	.x	.x	PROPN
ap-948	50	11	�	�	PROPN
ap-948	50	12	�	�	PROPN
ap-948	50	13	i	i	PROPN
ap-948	50	14	d	d	X
ap-948	50	15	(	(	PUNCT
ap-948	50	16	4	4	X
ap-948	50	17	)	)	PUNCT
ap-948	50	18	this	this	DET
ap-948	50	19	measure	measure	NOUN
ap-948	50	20	is	be	AUX
ap-948	50	21	not	not	PART
ap-948	50	22	suitable	suitable	ADJ
ap-948	50	23	for	for	ADP
ap-948	50	24	detecting	detect	VERB
ap-948	50	25	the	the	DET
ap-948	50	26	maximum	maximum	NOUN
ap-948	50	27	over	over	ADP
ap-948	50	28	scales	scale	NOUN
ap-948	50	29	in	in	ADP
ap-948	50	30	a	a	DET
ap-948	50	31	scale	scale	NOUN
ap-948	50	32	-	-	PUNCT
ap-948	50	33	space	space	NOUN
ap-948	50	34	representation	representation	NOUN
ap-948	50	35	.	.	PUNCT
ap-948	51	1	thus	thus	ADV
ap-948	51	2	the	the	DET
ap-948	51	3	normalized	normalize	VERB
ap-948	51	4	laplacian	laplacian	ADJ
ap-948	51	5	-	-	PUNCT
ap-948	51	6	of	of	ADP
ap-948	51	7	-	-	PUNCT
ap-948	51	8	gaussian	gaussian	NOUN
ap-948	51	9	is	be	AUX
ap-948	51	10	used	use	VERB
ap-948	51	11	for	for	ADP
ap-948	51	12	automatic	automatic	ADJ
ap-948	51	13	scale	scale	NOUN
ap-948	51	14	selection	selection	NOUN
ap-948	51	15	[	[	X
ap-948	51	16	5	5	NUM
ap-948	51	17	]	]	PUNCT
ap-948	51	18	.	.	PUNCT
ap-948	52	1	log	log	PROPN
ap-948	52	2	s	s	PART
ap-948	52	3	l	l	NOUN
ap-948	52	4	ln	ln	ADJ
ap-948	52	5	n	n	ADV
ap-948	52	6	xx	xx	NUM
ap-948	52	7	n	n	PROPN
ap-948	52	8	yy	yy	PROPN
ap-948	52	9	n	n	CCONJ
ap-948	52	10	(	(	PUNCT
ap-948	52	11	,	,	PUNCT
ap-948	52	12	)	)	PUNCT
ap-948	52	13	(	(	PUNCT
ap-948	52	14	,	,	PUNCT
ap-948	52	15	)	)	PUNCT
ap-948	52	16	(	(	PUNCT
ap-948	52	17	,	,	PUNCT
ap-948	52	18	)	)	PUNCT
ap-948	52	19	x	x	SYM
ap-948	52	20	x	x	SYM
ap-948	52	21	x	x	X
ap-948	52	22	�	�	PROPN
ap-948	52	23	�	�	PROPN
ap-948	52	24	�	�	PROPN
ap-948	52	25	�	�	PROPN
ap-948	52	26	�	�	PROPN
ap-948	52	27	2	2	NUM
ap-948	52	28	2	2	NUM
ap-948	52	29	2	2	NUM
ap-948	52	30	(	(	PUNCT
ap-948	52	31	5	5	NUM
ap-948	52	32	)	)	PUNCT
ap-948	52	33	the	the	DET
ap-948	52	34	region	region	NOUN
ap-948	52	35	around	around	ADP
ap-948	52	36	interest	interest	NOUN
ap-948	52	37	points	point	NOUN
ap-948	52	38	is	be	AUX
ap-948	52	39	described	describe	VERB
ap-948	52	40	using	use	VERB
ap-948	52	41	local	local	ADJ
ap-948	52	42	descriptors	descriptor	NOUN
ap-948	52	43	.	.	PUNCT
ap-948	53	1	recently	recently	ADV
ap-948	53	2	,	,	PUNCT
ap-948	53	3	several	several	ADJ
ap-948	53	4	descriptors	descriptor	NOUN
ap-948	53	5	have	have	AUX
ap-948	53	6	been	be	AUX
ap-948	53	7	developed	develop	VERB
ap-948	53	8	[	[	PUNCT
ap-948	53	9	8	8	NUM
ap-948	53	10	]	]	PUNCT
ap-948	53	11	.	.	PUNCT
ap-948	54	1	this	this	DET
ap-948	54	2	paper	paper	NOUN
ap-948	54	3	uses	use	VERB
ap-948	54	4	the	the	DET
ap-948	54	5	sift	sift	ADJ
ap-948	54	6	descriptor	descriptor	NOUN
ap-948	54	7	based	base	VERB
ap-948	54	8	on	on	ADP
ap-948	54	9	the	the	DET
ap-948	54	10	gradient	gradient	ADJ
ap-948	54	11	distribution	distribution	NOUN
ap-948	54	12	in	in	ADP
ap-948	54	13	the	the	DET
ap-948	54	14	detected	detect	VERB
ap-948	54	15	region	region	NOUN
ap-948	54	16	around	around	ADP
ap-948	54	17	the	the	DET
ap-948	54	18	interest	interest	NOUN
ap-948	54	19	point	point	NOUN
ap-948	54	20	[	[	X
ap-948	54	21	6	6	NUM
ap-948	54	22	]	]	PUNCT
ap-948	54	23	.	.	PUNCT
ap-948	55	1	the	the	DET
ap-948	55	2	size	size	NOUN
ap-948	55	3	of	of	ADP
ap-948	55	4	the	the	DET
ap-948	55	5	region	region	NOUN
ap-948	55	6	being	be	AUX
ap-948	55	7	described	describe	VERB
ap-948	55	8	depends	depend	VERB
ap-948	55	9	on	on	ADP
ap-948	55	10	the	the	DET
ap-948	55	11	detection	detection	NOUN
ap-948	55	12	scale	scale	NOUN
ap-948	55	13	�	�	PROPN
ap-948	55	14	.	.	PUNCT
ap-948	56	1	thus	thus	ADV
ap-948	56	2	a	a	DET
ap-948	56	3	scale	scale	NOUN
ap-948	56	4	invariant	invariant	ADJ
ap-948	56	5	description	description	NOUN
ap-948	56	6	is	be	AUX
ap-948	56	7	obtained	obtain	VERB
ap-948	56	8	.	.	PUNCT
ap-948	57	1	the	the	DET
ap-948	57	2	resulting	result	VERB
ap-948	57	3	descriptor	descriptor	NOUN
ap-948	57	4	maps	map	NOUN
ap-948	57	5	each	each	DET
ap-948	57	6	interest	interest	NOUN
ap-948	57	7	point	point	NOUN
ap-948	57	8	into	into	ADP
ap-948	57	9	a	a	DET
ap-948	57	10	128	128	NUM
ap-948	57	11	-	-	PUNCT
ap-948	57	12	dimensional	dimensional	ADJ
ap-948	57	13	vector	vector	NOUN
ap-948	57	14	m.	m.	NOUN
ap-948	57	15	in	in	ADP
ap-948	57	16	addition	addition	NOUN
ap-948	57	17	,	,	PUNCT
ap-948	57	18	a	a	DET
ap-948	57	19	dominant	dominant	ADJ
ap-948	57	20	orientation	orientation	NOUN
ap-948	57	21	is	be	AUX
ap-948	57	22	calculated	calculate	VERB
ap-948	57	23	for	for	ADP
ap-948	57	24	each	each	DET
ap-948	57	25	interest	interest	NOUN
ap-948	57	26	point	point	NOUN
ap-948	57	27	and	and	CCONJ
ap-948	57	28	descriptors	descriptor	NOUN
ap-948	57	29	are	be	AUX
ap-948	57	30	calculated	calculate	VERB
ap-948	57	31	on	on	ADP
ap-948	57	32	the	the	DET
ap-948	57	33	basis	basis	NOUN
ap-948	57	34	of	of	ADP
ap-948	57	35	this	this	DET
ap-948	57	36	angle	angle	NOUN
ap-948	57	37	to	to	PART
ap-948	57	38	gain	gain	VERB
ap-948	57	39	rotation	rotation	NOUN
ap-948	57	40	invariance	invariance	NOUN
ap-948	57	41	.	.	PUNCT
ap-948	58	1	3	3	NUM
ap-948	58	2	feature	feature	NOUN
ap-948	58	3	reduction	reduction	NOUN
ap-948	58	4	and	and	CCONJ
ap-948	58	5	selection	selection	NOUN
ap-948	58	6	in	in	ADP
ap-948	58	7	this	this	DET
ap-948	58	8	section	section	NOUN
ap-948	58	9	,	,	PUNCT
ap-948	58	10	dimensionality	dimensionality	NOUN
ap-948	58	11	reduction	reduction	NOUN
ap-948	58	12	and	and	CCONJ
ap-948	58	13	feature	feature	NOUN
ap-948	58	14	selection	selection	NOUN
ap-948	58	15	are	be	AUX
ap-948	58	16	explained	explain	VERB
ap-948	58	17	.	.	PUNCT
ap-948	59	1	the	the	DET
ap-948	59	2	karhunen	karhunen	NOUN
ap-948	59	3	-	-	PUNCT
ap-948	59	4	loeve	loeve	NOUN
ap-948	59	5	transform	transform	NOUN
ap-948	59	6	is	be	AUX
ap-948	59	7	applied	apply	VERB
ap-948	59	8	to	to	PART
ap-948	59	9	reduce	reduce	VERB
ap-948	59	10	the	the	DET
ap-948	59	11	dimensionality	dimensionality	NOUN
ap-948	59	12	.	.	PUNCT
ap-948	60	1	then	then	ADV
ap-948	60	2	the	the	DET
ap-948	60	3	clustering	clustering	ADJ
ap-948	60	4	process	process	NOUN
ap-948	60	5	and	and	CCONJ
ap-948	60	6	the	the	DET
ap-948	60	7	cluster	cluster	NOUN
ap-948	60	8	analysis	analysis	NOUN
ap-948	60	9	is	be	AUX
ap-948	60	10	described	describe	VERB
ap-948	60	11	.	.	PUNCT
ap-948	61	1	3.1	3.1	NUM
ap-948	61	2	karhunen	karhunen	NOUN
ap-948	61	3	-	-	PUNCT
ap-948	61	4	loeve	loeve	NOUN
ap-948	61	5	transform	transform	NOUN
ap-948	61	6	it	it	PRON
ap-948	61	7	is	be	AUX
ap-948	61	8	a	a	DET
ap-948	61	9	well	well	ADV
ap-948	61	10	-	-	PUNCT
ap-948	61	11	known	know	VERB
ap-948	61	12	problem	problem	NOUN
ap-948	61	13	in	in	ADP
ap-948	61	14	the	the	DET
ap-948	61	15	machine	machine	NOUN
ap-948	61	16	learning	learn	VERB
ap-948	61	17	domain	domain	NOUN
ap-948	61	18	that	that	SCONJ
ap-948	61	19	the	the	DET
ap-948	61	20	number	number	NOUN
ap-948	61	21	of	of	ADP
ap-948	61	22	required	require	VERB
ap-948	61	23	training	training	NOUN
ap-948	61	24	samples	sample	NOUN
ap-948	61	25	increases	increase	NOUN
ap-948	61	26	exponentially	exponentially	ADV
ap-948	61	27	with	with	ADP
ap-948	61	28	the	the	DET
ap-948	61	29	dimensionality	dimensionality	NOUN
ap-948	61	30	of	of	ADP
ap-948	61	31	the	the	DET
ap-948	61	32	feature	feature	NOUN
ap-948	61	33	space	space	NOUN
ap-948	61	34	.	.	PUNCT
ap-948	62	1	this	this	PRON
ap-948	62	2	is	be	AUX
ap-948	62	3	called	call	VERB
ap-948	62	4	the	the	DET
ap-948	62	5	“	"	PUNCT
ap-948	62	6	curse	curse	NOUN
ap-948	62	7	of	of	ADP
ap-948	62	8	dimensionality	dimensionality	NOUN
ap-948	62	9	”	"	PUNCT
ap-948	63	1	[	[	X
ap-948	63	2	1	1	NUM
ap-948	63	3	]	]	PUNCT
ap-948	63	4	.	.	PUNCT
ap-948	64	1	since	since	SCONJ
ap-948	64	2	the	the	DET
ap-948	64	3	sift	sift	NOUN
ap-948	64	4	descriptor	descriptor	NOUN
ap-948	64	5	forms	form	VERB
ap-948	64	6	a	a	DET
ap-948	64	7	128	128	NUM
ap-948	64	8	dimensional	dimensional	ADJ
ap-948	64	9	vector	vector	NOUN
ap-948	64	10	,	,	PUNCT
ap-948	64	11	the	the	DET
ap-948	64	12	karhunen	karhunen	NOUN
ap-948	64	13	-	-	PUNCT
ap-948	64	14	loeve	loeve	NOUN
ap-948	64	15	transform	transform	NOUN
ap-948	64	16	is	be	AUX
ap-948	64	17	applied	apply	VERB
ap-948	64	18	to	to	PART
ap-948	64	19	reduce	reduce	VERB
ap-948	64	20	dimensionality	dimensionality	NOUN
ap-948	64	21	.	.	PUNCT
ap-948	65	1	positive	positive	ADJ
ap-948	65	2	side	side	NOUN
ap-948	65	3	effects	effect	NOUN
ap-948	65	4	of	of	ADP
ap-948	65	5	this	this	PRON
ap-948	65	6	are	be	AUX
ap-948	65	7	a	a	DET
ap-948	65	8	reduction	reduction	NOUN
ap-948	65	9	in	in	ADP
ap-948	65	10	computing	compute	VERB
ap-948	65	11	time	time	NOUN
ap-948	65	12	for	for	ADP
ap-948	65	13	the	the	DET
ap-948	65	14	classification	classification	NOUN
ap-948	65	15	and	and	CCONJ
ap-948	65	16	the	the	DET
ap-948	65	17	fact	fact	NOUN
ap-948	65	18	that	that	SCONJ
ap-948	65	19	the	the	DET
ap-948	65	20	elements	element	NOUN
ap-948	65	21	of	of	ADP
ap-948	65	22	the	the	DET
ap-948	65	23	feature	feature	NOUN
ap-948	65	24	vector	vector	NOUN
ap-948	65	25	m	m	AUX
ap-948	65	26	become	become	VERB
ap-948	65	27	uncorrelated	uncorrelated	ADJ
ap-948	65	28	.	.	PUNCT
ap-948	66	1	hence	hence	ADV
ap-948	66	2	the	the	DET
ap-948	66	3	covariance	covariance	NOUN
ap-948	66	4	matrix	matrix	NOUN
ap-948	66	5	�	�	PROPN
ap-948	66	6	of	of	ADP
ap-948	66	7	n	n	NOUN
ap-948	66	8	training	train	VERB
ap-948	66	9	feature	feature	NOUN
ap-948	66	10	vectors	vector	NOUN
ap-948	66	11	m	m	VERB
ap-948	66	12	is	be	AUX
ap-948	66	13	used	use	VERB
ap-948	66	14	to	to	PART
ap-948	66	15	find	find	VERB
ap-948	66	16	a	a	DET
ap-948	66	17	linear	linear	ADJ
ap-948	66	18	subspace	subspace	NOUN
ap-948	66	19	for	for	ADP
ap-948	66	20	better	well	ADJ
ap-948	66	21	representation	representation	NOUN
ap-948	66	22	.	.	PUNCT
ap-948	67	1	�	�	PROPN
ap-948	67	2	�	�	PROPN
ap-948	67	3	�	�	PROPN
ap-948	67	4	�	�	PROPN
ap-948	67	5	�	�	PROPN
ap-948	67	6	�	�	PROPN
ap-948	67	7	�	�	PROPN
ap-948	67	8	�	�	PROPN
ap-948	67	9	�	�	PROPN
ap-948	67	10	�	�	PROPN
ap-948	67	11	�	�	PROPN
ap-948	67	12	1	1	NUM
ap-948	67	13	1	1	NUM
ap-948	67	14	1	1	NUM
ap-948	67	15	1	1	NUM
ap-948	67	16	1	1	NUM
ap-948	67	17	n	n	NUM
ap-948	67	18	n	n	NOUN
ap-948	67	19	i	i	PRON
ap-948	67	20	i	i	PRON
ap-948	67	21	n	n	VERB
ap-948	68	1	i	i	PRON
ap-948	68	2	i	i	PRON
ap-948	68	3	t	t	VERB
ap-948	69	1	i	i	PRON
ap-948	70	1	n	n	VERB
ap-948	70	2	m	m	VERB
ap-948	70	3	m	m	VERB
ap-948	70	4	m	m	ADJ
ap-948	70	5	(	(	PUNCT
ap-948	70	6	)	)	PUNCT
ap-948	70	7	(	(	PUNCT
ap-948	70	8	)	)	PUNCT
ap-948	70	9	therefore	therefore	ADV
ap-948	70	10	the	the	DET
ap-948	70	11	covariance	covariance	NOUN
ap-948	70	12	matrix	matrix	NOUN
ap-948	70	13	�	�	PROPN
ap-948	70	14	has	have	VERB
ap-948	70	15	to	to	PART
ap-948	70	16	be	be	AUX
ap-948	70	17	orthogonalized	orthogonalize	VERB
ap-948	70	18	by	by	ADP
ap-948	70	19	an	an	DET
ap-948	70	20	eigenvector	eigenvector	NOUN
ap-948	70	21	transform	transform	NOUN
ap-948	70	22	:	:	PUNCT
ap-948	70	23	�	�	PROPN
ap-948	70	24	�	�	PROPN
ap-948	70	25	�	�	PROPN
ap-948	70	26	t	t	PROPN
ap-948	70	27	n	n	CCONJ
ap-948	70	28	�	�	PROPN
ap-948	70	29	�	�	PROPN
ap-948	70	30	�	�	PROPN
ap-948	70	31	�	�	PROPN
ap-948	70	32	�	�	PROPN
ap-948	70	33	�	�	PROPN
ap-948	70	34	�	�	PROPN
ap-948	70	35	�	�	PROPN
ap-948	70	36	�	�	PROPN
ap-948	70	37	�	�	PROPN
ap-948	70	38	�	�	PROPN
ap-948	70	39	�	�	PROPN
ap-948	70	40	�	�	PROPN
ap-948	70	41	�	�	PROPN
ap-948	70	42	1	1	NUM
ap-948	70	43	2	2	NUM
ap-948	70	44	0	0	NUM
ap-948	70	45	0	0	NUM
ap-948	70	46	0	0	NUM
ap-948	70	47	0	0	NUM
ap-948	70	48	0	0	SYM
ap-948	70	49	0	0	NUM
ap-948	70	50	�	�	PROPN
ap-948	70	51	�	�	PROPN
ap-948	70	52	�	�	PROPN
ap-948	70	53	�	�	PROPN
ap-948	70	54	�	�	PROPN
ap-948	70	55	�	�	PROPN
ap-948	70	56	�	�	PROPN
ap-948	70	57	.	.	PUNCT
ap-948	71	1	(	(	PUNCT
ap-948	71	2	6	6	X
ap-948	71	3	)	)	PUNCT
ap-948	71	4	the	the	DET
ap-948	71	5	columns	column	NOUN
ap-948	71	6	of	of	ADP
ap-948	71	7	transform	transform	NOUN
ap-948	71	8	matrix	matrix	NOUN
ap-948	71	9	�	�	NOUN
ap-948	71	10	are	be	AUX
ap-948	71	11	the	the	DET
ap-948	71	12	eigenvectors	eigenvector	NOUN
ap-948	71	13	of	of	ADP
ap-948	71	14	�	�	PROPN
ap-948	71	15	.	.	PUNCT
ap-948	72	1	these	these	DET
ap-948	72	2	eigenvectors	eigenvector	NOUN
ap-948	72	3	are	be	AUX
ap-948	72	4	orthonormal	orthonormal	ADJ
ap-948	72	5	and	and	CCONJ
ap-948	72	6	represent	represent	VERB
ap-948	72	7	the	the	DET
ap-948	72	8	basis	basis	NOUN
ap-948	72	9	vectors	vector	NOUN
ap-948	72	10	.	.	PUNCT
ap-948	73	1	the	the	DET
ap-948	73	2	use	use	NOUN
ap-948	73	3	of	of	ADP
ap-948	73	4	only	only	ADV
ap-948	73	5	a	a	DET
ap-948	73	6	subset	subset	NOUN
ap-948	73	7	of	of	ADP
ap-948	73	8	these	these	DET
ap-948	73	9	eigenvectors	eigenvector	NOUN
ap-948	73	10	as	as	SCONJ
ap-948	73	11	transformation	transformation	NOUN
ap-948	73	12	matrix	matrix	NOUN
ap-948	73	13	�	�	PROPN
ap-948	73	14	t	t	PROPN
ap-948	73	15	leads	lead	VERB
ap-948	73	16	to	to	ADP
ap-948	73	17	reduced	reduce	VERB
ap-948	73	18	dimensionality	dimensionality	NOUN
ap-948	73	19	of	of	ADP
ap-948	73	20	the	the	DET
ap-948	73	21	feature	feature	NOUN
ap-948	73	22	space	space	NOUN
ap-948	73	23	[	[	X
ap-948	73	24	2	2	NUM
ap-948	73	25	]	]	PUNCT
ap-948	73	26	:	:	PUNCT
ap-948	73	27	f	f	PROPN
ap-948	73	28	m	m	PROPN
ap-948	73	29	�	�	PROPN
ap-948	73	30	�	�	PROPN
ap-948	73	31	�	�	PROPN
ap-948	73	32	�	�	PROPN
ap-948	73	33	(	(	PUNCT
ap-948	73	34	)	)	PUNCT
ap-948	73	35	�	�	PROPN
ap-948	73	36	�	�	PROPN
ap-948	73	37	t	t	PROPN
ap-948	73	38	(	(	PUNCT
ap-948	73	39	7	7	NUM
ap-948	73	40	)	)	PUNCT
ap-948	73	41	3.2	3.2	NUM
ap-948	73	42	clustering	cluster	VERB
ap-948	73	43	in	in	ADP
ap-948	73	44	contrast	contrast	NOUN
ap-948	73	45	to	to	ADP
ap-948	73	46	knowledge	knowledge	NOUN
ap-948	73	47	-	-	PUNCT
ap-948	73	48	based	base	VERB
ap-948	73	49	methods	method	NOUN
ap-948	73	50	,	,	PUNCT
ap-948	73	51	the	the	DET
ap-948	73	52	exact	exact	ADJ
ap-948	73	53	facial	facial	ADJ
ap-948	73	54	features	feature	NOUN
ap-948	73	55	are	be	AUX
ap-948	73	56	not	not	PART
ap-948	73	57	defined	define	VERB
ap-948	73	58	beforehand	beforehand	ADV
ap-948	73	59	.	.	PUNCT
ap-948	74	1	for	for	ADP
ap-948	74	2	clustering	cluster	VERB
ap-948	74	3	the	the	DET
ap-948	74	4	training	training	NOUN
ap-948	74	5	feature	feature	NOUN
ap-948	74	6	vectors	vector	NOUN
ap-948	74	7	f	f	PROPN
ap-948	74	8	in	in	ADP
ap-948	74	9	the	the	DET
ap-948	74	10	reduced	reduce	VERB
ap-948	74	11	feature	feature	NOUN
ap-948	74	12	space	space	NOUN
ap-948	74	13	,	,	PUNCT
ap-948	74	14	we	we	PRON
ap-948	74	15	need	need	VERB
ap-948	74	16	,	,	PUNCT
ap-948	74	17	however	however	ADV
ap-948	74	18	,	,	PUNCT
ap-948	74	19	to	to	PART
ap-948	74	20	estimate	estimate	VERB
ap-948	74	21	their	their	PRON
ap-948	74	22	number	number	NOUN
ap-948	74	23	.	.	PUNCT
ap-948	75	1	a	a	DET
ap-948	75	2	single	single	ADJ
ap-948	75	3	cluster	cluster	NOUN
ap-948	75	4	represents	represent	VERB
ap-948	75	5	a	a	DET
ap-948	75	6	distribution	distribution	NOUN
ap-948	75	7	of	of	ADP
ap-948	75	8	a	a	DET
ap-948	75	9	number	number	NOUN
ap-948	75	10	of	of	ADP
ap-948	75	11	feature	feature	NOUN
ap-948	75	12	vectors	vector	NOUN
ap-948	75	13	f	f	PROPN
ap-948	75	14	that	that	PRON
ap-948	75	15	belongs	belong	VERB
ap-948	75	16	to	to	ADP
ap-948	75	17	the	the	DET
ap-948	75	18	same	same	ADJ
ap-948	75	19	facial	facial	ADJ
ap-948	75	20	feature	feature	NOUN
ap-948	75	21	.	.	PUNCT
ap-948	76	1	the	the	DET
ap-948	76	2	number	number	NOUN
ap-948	76	3	of	of	ADP
ap-948	76	4	clusters	cluster	NOUN
ap-948	76	5	must	must	AUX
ap-948	76	6	therefore	therefore	ADV
ap-948	76	7	be	be	AUX
ap-948	76	8	big	big	ADJ
ap-948	76	9	enough	enough	ADV
ap-948	76	10	to	to	PART
ap-948	76	11	offer	offer	VERB
ap-948	76	12	at	at	ADV
ap-948	76	13	least	least	ADV
ap-948	76	14	one	one	NUM
ap-948	76	15	cluster	cluster	NOUN
ap-948	76	16	centroid	centroid	NOUN
ap-948	76	17	per	per	ADP
ap-948	76	18	facial	facial	ADJ
ap-948	76	19	feature	feature	NOUN
ap-948	76	20	.	.	PUNCT
ap-948	77	1	choosing	choose	VERB
ap-948	77	2	too	too	ADV
ap-948	77	3	many	many	ADJ
ap-948	77	4	clusters	cluster	NOUN
ap-948	77	5	,	,	PUNCT
ap-948	77	6	however	however	ADV
ap-948	77	7	,	,	PUNCT
ap-948	77	8	leads	lead	VERB
ap-948	77	9	to	to	ADP
ap-948	77	10	overtraining	overtraining	NOUN
ap-948	77	11	,	,	PUNCT
ap-948	77	12	i.e.	i.e.	X
ap-948	77	13	a	a	DET
ap-948	77	14	single	single	ADJ
ap-948	77	15	facial	facial	ADJ
ap-948	77	16	feature	feature	NOUN
ap-948	77	17	will	will	AUX
ap-948	77	18	be	be	AUX
ap-948	77	19	represented	represent	VERB
ap-948	77	20	by	by	ADP
ap-948	77	21	a	a	DET
ap-948	77	22	multitude	multitude	NOUN
ap-948	77	23	of	of	ADP
ap-948	77	24	cluster	cluster	NOUN
ap-948	77	25	centroids	centroid	NOUN
ap-948	77	26	that	that	PRON
ap-948	77	27	are	be	AUX
ap-948	77	28	adapted	adapt	VERB
ap-948	77	29	to	to	ADP
ap-948	77	30	the	the	DET
ap-948	77	31	training	training	NOUN
ap-948	77	32	set	set	VERB
ap-948	77	33	instead	instead	ADV
ap-948	77	34	of	of	ADP
ap-948	77	35	generalizing	generalize	VERB
ap-948	77	36	a	a	DET
ap-948	77	37	given	give	VERB
ap-948	77	38	feature	feature	NOUN
ap-948	77	39	.	.	PUNCT
ap-948	78	1	hence	hence	ADV
ap-948	78	2	an	an	DET
ap-948	78	3	evaluation	evaluation	NOUN
ap-948	78	4	with	with	ADP
ap-948	78	5	different	different	ADJ
ap-948	78	6	numbers	number	NOUN
ap-948	78	7	of	of	ADP
ap-948	78	8	clusters	cluster	NOUN
ap-948	78	9	was	be	AUX
ap-948	78	10	conducted	conduct	VERB
ap-948	78	11	(	(	PUNCT
ap-948	78	12	see	see	VERB
ap-948	78	13	section	section	NOUN
ap-948	78	14	4	4	NUM
ap-948	78	15	)	)	PUNCT
ap-948	78	16	to	to	PART
ap-948	78	17	find	find	VERB
ap-948	78	18	the	the	DET
ap-948	78	19	optimum	optimum	ADJ
ap-948	78	20	value	value	NOUN
ap-948	78	21	ex	ex	X
ap-948	78	22	post	post	NOUN
ap-948	78	23	based	base	VERB
ap-948	78	24	on	on	ADP
ap-948	78	25	the	the	DET
ap-948	78	26	classifier	classifier	NOUN
ap-948	78	27	performance	performance	NOUN
ap-948	78	28	.	.	PUNCT
ap-948	79	1	a	a	DET
ap-948	79	2	detailed	detailed	ADJ
ap-948	79	3	description	description	NOUN
ap-948	79	4	of	of	ADP
ap-948	79	5	the	the	DET
ap-948	79	6	clustering	clustering	ADJ
ap-948	79	7	process	process	NOUN
ap-948	79	8	is	be	AUX
ap-948	79	9	given	give	VERB
ap-948	79	10	in	in	ADP
ap-948	79	11	the	the	DET
ap-948	79	12	following	follow	VERB
ap-948	79	13	paragraphs	paragraph	NOUN
ap-948	79	14	.	.	PUNCT
ap-948	80	1	first	first	ADV
ap-948	80	2	of	of	ADP
ap-948	80	3	all	all	PRON
ap-948	80	4	,	,	PUNCT
ap-948	80	5	the	the	DET
ap-948	80	6	training	training	NOUN
ap-948	80	7	feature	feature	NOUN
ap-948	80	8	vectors	vector	NOUN
ap-948	80	9	f	f	PRON
ap-948	80	10	are	be	AUX
ap-948	80	11	split	split	VERB
ap-948	80	12	up	up	ADP
ap-948	80	13	into	into	ADP
ap-948	80	14	two	two	NUM
ap-948	80	15	subsets	subset	NOUN
ap-948	80	16	.	.	PUNCT
ap-948	81	1	the	the	DET
ap-948	81	2	first	first	ADJ
ap-948	81	3	subset	subset	NOUN
ap-948	81	4	contains	contain	VERB
ap-948	81	5	all	all	DET
ap-948	81	6	feature	feature	NOUN
ap-948	81	7	vectors	vector	NOUN
ap-948	81	8	describing	describe	VERB
ap-948	81	9	the	the	DET
ap-948	81	10	facial	facial	ADJ
ap-948	81	11	features	feature	NOUN
ap-948	81	12	(	(	PUNCT
ap-948	81	13	class	class	NOUN
ap-948	81	14	a	a	DET
ap-948	81	15	�	�	PROPN
ap-948	81	16	1	1	NUM
ap-948	81	17	)	)	PUNCT
ap-948	81	18	.	.	PUNCT
ap-948	82	1	in	in	ADP
ap-948	82	2	the	the	DET
ap-948	82	3	second	second	ADJ
ap-948	82	4	subset	subset	NOUN
ap-948	82	5	all	all	DET
ap-948	82	6	feature	feature	NOUN
ap-948	82	7	vectors	vector	NOUN
ap-948	82	8	describing	describe	VERB
ap-948	82	9	the	the	DET
ap-948	82	10	rest	rest	NOUN
ap-948	82	11	of	of	ADP
ap-948	82	12	the	the	DET
ap-948	82	13	image	image	NOUN
ap-948	82	14	are	be	AUX
ap-948	82	15	aggregated	aggregate	VERB
ap-948	82	16	(	(	PUNCT
ap-948	82	17	class	class	NOUN
ap-948	82	18	a	a	DET
ap-948	82	19	�	�	PROPN
ap-948	82	20	0	0	NUM
ap-948	82	21	):	):	PUNCT
ap-948	82	22	�	�	PROPN
ap-948	82	23	�	�	PROPN
ap-948	82	24	�	�	PROPN
ap-948	82	25	�	�	PROPN
ap-948	82	26	f	f	PROPN
ap-948	82	27	a	a	DET
ap-948	82	28	f	f	PROPN
ap-948	82	29	aface	aface	PROPN
ap-948	82	30	non	non	PROPN
ap-948	82	31	face	face	NOUN
ap-948	82	32	�	�	PROPN
ap-948	82	33	�	�	PROPN
ap-948	82	34	�	�	PROPN
ap-948	82	35	�	�	PROPN
ap-948	82	36	�	�	PROPN
ap-948	82	37	�	�	PROPN
ap-948	82	38	f	f	PROPN
ap-948	82	39	f1	f1	PROPN
ap-948	82	40	0	0	NUM
ap-948	82	41	,	,	PUNCT
ap-948	82	42	(	(	PUNCT
ap-948	82	43	8)	8)	NUM
ap-948	82	44	each	each	DET
ap-948	82	45	subset	subset	NOUN
ap-948	82	46	is	be	AUX
ap-948	82	47	clustered	cluster	VERB
ap-948	82	48	with	with	ADP
ap-948	82	49	the	the	DET
ap-948	82	50	k	k	NOUN
ap-948	82	51	-	-	PUNCT
ap-948	82	52	means	means	NOUN
ap-948	82	53	algorithm	algorithm	NOUN
ap-948	82	54	.	.	PUNCT
ap-948	83	1	thereby	thereby	ADV
ap-948	83	2	the	the	DET
ap-948	83	3	number	number	NOUN
ap-948	83	4	of	of	ADP
ap-948	83	5	clusters	cluster	NOUN
ap-948	83	6	in	in	ADP
ap-948	83	7	each	each	DET
ap-948	83	8	subset	subset	NOUN
ap-948	83	9	is	be	AUX
ap-948	83	10	kept	keep	VERB
ap-948	83	11	proportional	proportional	ADJ
ap-948	83	12	to	to	ADP
ap-948	83	13	the	the	DET
ap-948	83	14	respective	respective	ADJ
ap-948	83	15	number	number	NOUN
ap-948	83	16	of	of	ADP
ap-948	83	17	features	feature	NOUN
ap-948	83	18	f.	f.	PROPN
ap-948	83	19	in	in	ADP
ap-948	83	20	other	other	ADJ
ap-948	83	21	words	word	NOUN
ap-948	83	22	the	the	DET
ap-948	83	23	average	average	ADJ
ap-948	83	24	number	number	NOUN
ap-948	83	25	of	of	ADP
ap-948	83	26	features	feature	NOUN
ap-948	83	27	represented	represent	VERB
ap-948	83	28	by	by	ADP
ap-948	83	29	each	each	DET
ap-948	83	30	cluster	cluster	NOUN
ap-948	83	31	is	be	AUX
ap-948	83	32	chosen	choose	VERB
ap-948	83	33	to	to	PART
ap-948	83	34	be	be	AUX
ap-948	83	35	identical	identical	ADJ
ap-948	83	36	for	for	ADP
ap-948	83	37	both	both	DET
ap-948	83	38	subsets	subset	NOUN
ap-948	83	39	.	.	PUNCT
ap-948	84	1	3.3	3.3	NUM
ap-948	84	2	cluster	cluster	NOUN
ap-948	84	3	analysis	analysis	NOUN
ap-948	84	4	a	a	DET
ap-948	84	5	cluster	cluster	NOUN
ap-948	84	6	analysis	analysis	NOUN
ap-948	84	7	is	be	AUX
ap-948	84	8	applied	apply	VERB
ap-948	84	9	to	to	ADP
ap-948	84	10	fface	fface	NOUN
ap-948	84	11	to	to	PART
ap-948	84	12	find	find	VERB
ap-948	84	13	the	the	DET
ap-948	84	14	most	most	ADV
ap-948	84	15	characteristic	characteristic	ADJ
ap-948	84	16	clusters	cluster	NOUN
ap-948	84	17	of	of	ADP
ap-948	84	18	facial	facial	ADJ
ap-948	84	19	features	feature	NOUN
ap-948	84	20	.	.	PUNCT
ap-948	85	1	the	the	DET
ap-948	85	2	cluster	cluster	NOUN
ap-948	85	3	precision	precision	NOUN
ap-948	85	4	introduced	introduce	VERB
ap-948	85	5	in	in	ADP
ap-948	85	6	[	[	X
ap-948	85	7	9	9	NUM
ap-948	85	8	]	]	PUNCT
ap-948	85	9	is	be	AUX
ap-948	85	10	a	a	DET
ap-948	85	11	measure	measure	NOUN
ap-948	85	12	of	of	ADP
ap-948	85	13	the	the	DET
ap-948	85	14	representativeness	representativeness	NOUN
ap-948	85	15	of	of	ADP
ap-948	85	16	a	a	DET
ap-948	85	17	cluster	cluster	NOUN
ap-948	85	18	:	:	PUNCT
ap-948	85	19	p	p	X
ap-948	85	20	f	f	PROPN
ap-948	85	21	f	f	PROPN
ap-948	85	22	fja	fja	NOUN
ap-948	85	23	ja	ja	PROPN
ap-948	85	24	ja	ja	PROPN
ap-948	85	25	ja	ja	PROPN
ap-948	85	26	�	�	PROPN
ap-948	85	27	�	�	PROPN
ap-948	85	28	�	�	PROPN
ap-948	85	29	�	�	PROPN
ap-948	85	30	#	#	NOUN
ap-948	85	31	#	#	NOUN
ap-948	85	32	#	#	SYM
ap-948	85	33	0	0	NUM
ap-948	85	34	1	1	NUM
ap-948	85	35	(	(	PUNCT
ap-948	85	36	9	9	NUM
ap-948	85	37	)	)	PUNCT
ap-948	85	38	4	4	NUM
ap-948	85	39	©	©	PROPN
ap-948	85	40	czech	czech	PROPN
ap-948	85	41	technical	technical	PROPN
ap-948	85	42	university	university	PROPN
ap-948	85	43	publishing	publishing	NOUN
ap-948	85	44	house	house	NOUN
ap-948	85	45	http://ctn.cvut.cz/ap/	http://ctn.cvut.cz/ap/	PROPN
ap-948	85	46	acta	acta	PROPN
ap-948	85	47	polytechnica	polytechnica	PROPN
ap-948	85	48	vol	vol	NOUN
ap-948	85	49	.	.	PUNCT
ap-948	86	1	47	47	NUM
ap-948	86	2	no	no	NOUN
ap-948	86	3	.	.	PUNCT
ap-948	87	1	4–5/2007	4–5/2007	NUM
ap-948	87	2	fig	fig	NOUN
ap-948	87	3	.	.	PUNCT
ap-948	88	1	2	2	NUM
ap-948	88	2	:	:	PUNCT
ap-948	88	3	cluster	cluster	NOUN
ap-948	88	4	j	j	PROPN
ap-948	88	5	represents	represent	VERB
ap-948	88	6	the	the	DET
ap-948	88	7	feature	feature	NOUN
ap-948	88	8	of	of	ADP
ap-948	88	9	class	class	NOUN
ap-948	88	10	a	a	DET
ap-948	88	11	�	�	PROPN
ap-948	88	12	0	0	NUM
ap-948	88	13	,	,	PUNCT
ap-948	88	14	1	1	NUM
ap-948	88	15	.	.	PUNCT
ap-948	89	1	the	the	DET
ap-948	89	2	cluster	cluster	NOUN
ap-948	89	3	precision	precision	NOUN
ap-948	89	4	is	be	AUX
ap-948	89	5	pja	pja	PROPN
ap-948	89	6	�	�	PROPN
ap-948	89	7	0	0	NUM
ap-948	89	8	�	�	PROPN
ap-948	89	9	0.67	0.67	NUM
ap-948	89	10	and	and	CCONJ
ap-948	89	11	pja	pja	PROPN
ap-948	89	12	�	�	PROPN
ap-948	89	13	1	1	NUM
ap-948	89	14	�	�	PROPN
ap-948	89	15	0.33	0.33	NUM
ap-948	89	16	.	.	PUNCT
ap-948	90	1	this	this	DET
ap-948	90	2	results	result	VERB
ap-948	90	3	in	in	ADP
ap-948	90	4	the	the	DET
ap-948	90	5	probability	probability	NOUN
ap-948	90	6	pja	pja	NOUN
ap-948	90	7	that	that	SCONJ
ap-948	90	8	a	a	DET
ap-948	90	9	feature	feature	NOUN
ap-948	90	10	of	of	ADP
ap-948	90	11	class	class	NOUN
ap-948	90	12	a	a	PRON
ap-948	90	13	is	be	AUX
ap-948	90	14	represented	represent	VERB
ap-948	90	15	by	by	ADP
ap-948	90	16	cluster	cluster	NOUN
ap-948	90	17	j.	j.	PROPN
ap-948	90	18	in	in	ADP
ap-948	90	19	order	order	NOUN
ap-948	90	20	to	to	PART
ap-948	90	21	determine	determine	VERB
ap-948	90	22	which	which	DET
ap-948	90	23	feature	feature	NOUN
ap-948	90	24	is	be	AUX
ap-948	90	25	represented	represent	VERB
ap-948	90	26	by	by	ADP
ap-948	90	27	which	which	DET
ap-948	90	28	cluster	cluster	NOUN
ap-948	90	29	,	,	PUNCT
ap-948	90	30	a	a	DET
ap-948	90	31	vector	vector	NOUN
ap-948	90	32	quantization	quantization	NOUN
ap-948	90	33	is	be	AUX
ap-948	90	34	applied	apply	VERB
ap-948	90	35	to	to	ADP
ap-948	90	36	all	all	DET
ap-948	90	37	training	train	VERB
ap-948	90	38	feature	feature	NOUN
ap-948	90	39	vectors	vector	NOUN
ap-948	90	40	f.	f.	PROPN
ap-948	90	41	thus	thus	ADV
ap-948	90	42	the	the	DET
ap-948	90	43	cluster	cluster	NOUN
ap-948	90	44	precision	precision	NOUN
ap-948	90	45	for	for	ADP
ap-948	90	46	each	each	DET
ap-948	90	47	cluster	cluster	NOUN
ap-948	90	48	is	be	AUX
ap-948	90	49	obtained	obtain	VERB
ap-948	90	50	.	.	PUNCT
ap-948	91	1	finally	finally	ADV
ap-948	91	2	only	only	ADV
ap-948	91	3	clusters	cluster	NOUN
ap-948	91	4	that	that	PRON
ap-948	91	5	represent	represent	VERB
ap-948	91	6	facial	facial	ADJ
ap-948	91	7	features	feature	NOUN
ap-948	91	8	(	(	PUNCT
ap-948	91	9	a	a	DET
ap-948	91	10	�	�	PROPN
ap-948	91	11	1	1	NUM
ap-948	91	12	)	)	PUNCT
ap-948	91	13	and	and	CCONJ
ap-948	91	14	whose	whose	DET
ap-948	91	15	cluster	cluster	NOUN
ap-948	91	16	precision	precision	NOUN
ap-948	91	17	is	be	AUX
ap-948	91	18	higher	high	ADJ
ap-948	91	19	than	than	SCONJ
ap-948	91	20	0.9	0.9	NUM
ap-948	91	21	are	be	AUX
ap-948	91	22	kept	keep	VERB
ap-948	91	23	to	to	PART
ap-948	91	24	form	form	VERB
ap-948	91	25	the	the	DET
ap-948	91	26	classifier	classifier	NOUN
ap-948	91	27	.	.	PUNCT
ap-948	92	1	in	in	ADP
ap-948	92	2	this	this	DET
ap-948	92	3	way	way	NOUN
ap-948	92	4	the	the	DET
ap-948	92	5	most	most	ADV
ap-948	92	6	distinctive	distinctive	ADJ
ap-948	92	7	clusters	cluster	NOUN
ap-948	92	8	representing	represent	VERB
ap-948	92	9	the	the	DET
ap-948	92	10	facial	facial	ADJ
ap-948	92	11	features	feature	NOUN
ap-948	92	12	in	in	ADP
ap-948	92	13	feature	feature	NOUN
ap-948	92	14	space	space	NOUN
ap-948	92	15	are	be	AUX
ap-948	92	16	determined	determine	VERB
ap-948	92	17	.	.	PUNCT
ap-948	93	1	in	in	ADP
ap-948	93	2	the	the	DET
ap-948	93	3	following	following	NOUN
ap-948	93	4	,	,	PUNCT
ap-948	93	5	these	these	DET
ap-948	93	6	clusters	cluster	NOUN
ap-948	93	7	will	will	AUX
ap-948	93	8	be	be	AUX
ap-948	93	9	denoted	denote	VERB
ap-948	93	10	as	as	ADP
ap-948	93	11	target	target	NOUN
ap-948	93	12	clusters	cluster	NOUN
ap-948	93	13	.	.	PUNCT
ap-948	94	1	4	4	NUM
ap-948	94	2	classifier	classifier	NOUN
ap-948	94	3	the	the	DET
ap-948	94	4	distances	distance	NOUN
ap-948	94	5	of	of	ADP
ap-948	94	6	the	the	DET
ap-948	94	7	feature	feature	NOUN
ap-948	94	8	vectors	vector	NOUN
ap-948	94	9	f	f	X
ap-948	94	10	to	to	ADP
ap-948	94	11	the	the	DET
ap-948	94	12	target	target	NOUN
ap-948	94	13	clusters	cluster	NOUN
ap-948	94	14	determined	determine	VERB
ap-948	94	15	in	in	ADP
ap-948	94	16	section	section	NOUN
ap-948	94	17	3.3	3.3	NUM
ap-948	94	18	,	,	PUNCT
ap-948	94	19	are	be	AUX
ap-948	94	20	a	a	DET
ap-948	94	21	measure	measure	NOUN
ap-948	94	22	of	of	ADP
ap-948	94	23	the	the	DET
ap-948	94	24	performance	performance	NOUN
ap-948	94	25	of	of	ADP
ap-948	94	26	the	the	DET
ap-948	94	27	classification	classification	NOUN
ap-948	94	28	process	process	NOUN
ap-948	94	29	.	.	PUNCT
ap-948	95	1	as	as	SCONJ
ap-948	95	2	shown	show	VERB
ap-948	95	3	in	in	ADP
ap-948	95	4	fig	fig	NOUN
ap-948	95	5	.	.	PUNCT
ap-948	96	1	3	3	NUM
ap-948	96	2	,	,	PUNCT
ap-948	96	3	the	the	DET
ap-948	96	4	training	training	NOUN
ap-948	96	5	features	feature	NOUN
ap-948	96	6	assigned	assign	VERB
ap-948	96	7	to	to	ADP
ap-948	96	8	a	a	DET
ap-948	96	9	face	face	NOUN
ap-948	96	10	(	(	PUNCT
ap-948	96	11	a	a	DET
ap-948	96	12	�	�	PROPN
ap-948	96	13	1	1	NUM
ap-948	96	14	)	)	PUNCT
ap-948	96	15	are	be	AUX
ap-948	96	16	closer	close	ADJ
ap-948	96	17	to	to	ADP
ap-948	96	18	the	the	DET
ap-948	96	19	target	target	NOUN
ap-948	96	20	cluster	cluster	NOUN
ap-948	96	21	than	than	ADP
ap-948	96	22	the	the	DET
ap-948	96	23	other	other	ADJ
ap-948	96	24	features	feature	NOUN
ap-948	96	25	(	(	PUNCT
ap-948	96	26	a	a	DET
ap-948	96	27	�	�	PROPN
ap-948	96	28	0	0	NUM
ap-948	96	29	)	)	PUNCT
ap-948	96	30	.	.	PUNCT
ap-948	97	1	a	a	DET
ap-948	97	2	threshold	threshold	NOUN
ap-948	97	3	can	can	AUX
ap-948	97	4	therefore	therefore	ADV
ap-948	97	5	be	be	AUX
ap-948	97	6	calculated	calculate	VERB
ap-948	97	7	from	from	ADP
ap-948	97	8	the	the	DET
ap-948	97	9	training	training	NOUN
ap-948	97	10	features	feature	NOUN
ap-948	97	11	for	for	ADP
ap-948	97	12	a	a	DET
ap-948	97	13	given	give	VERB
ap-948	97	14	false	false	ADJ
ap-948	97	15	-	-	PUNCT
ap-948	97	16	positive	positive	ADJ
ap-948	97	17	rate	rate	NOUN
ap-948	97	18	.	.	PUNCT
ap-948	98	1	in	in	ADP
ap-948	98	2	this	this	DET
ap-948	98	3	context	context	NOUN
ap-948	98	4	,	,	PUNCT
ap-948	98	5	false	false	ADJ
ap-948	98	6	-	-	PUNCT
ap-948	98	7	positive	positive	ADJ
ap-948	98	8	denotes	denote	NOUN
ap-948	98	9	all	all	DET
ap-948	98	10	features	feature	NOUN
ap-948	98	11	wrongly	wrongly	ADV
ap-948	98	12	classified	classify	VERB
ap-948	98	13	as	as	ADP
ap-948	98	14	facial	facial	ADJ
ap-948	98	15	features	feature	NOUN
ap-948	98	16	,	,	PUNCT
ap-948	98	17	whereas	whereas	SCONJ
ap-948	98	18	true	true	ADJ
ap-948	98	19	-	-	PUNCT
ap-948	98	20	positive	positive	ADJ
ap-948	98	21	denotes	denote	NOUN
ap-948	98	22	all	all	PRON
ap-948	98	23	features	feature	VERB
ap-948	98	24	correctly	correctly	ADV
ap-948	98	25	classified	classify	VERB
ap-948	98	26	as	as	ADP
ap-948	98	27	facial	facial	ADJ
ap-948	98	28	features	feature	NOUN
ap-948	98	29	.	.	PUNCT
ap-948	99	1	in	in	ADP
ap-948	99	2	this	this	DET
ap-948	99	3	way	way	NOUN
ap-948	99	4	a	a	DET
ap-948	99	5	classifier	classifier	NOUN
ap-948	99	6	is	be	AUX
ap-948	99	7	obtained	obtain	VERB
ap-948	99	8	.	.	PUNCT
ap-948	100	1	up	up	ADP
ap-948	100	2	this	this	DET
ap-948	100	3	point	point	NOUN
ap-948	100	4	,	,	PUNCT
ap-948	100	5	all	all	DET
ap-948	100	6	evaluations	evaluation	NOUN
ap-948	100	7	have	have	AUX
ap-948	100	8	been	be	AUX
ap-948	100	9	based	base	VERB
ap-948	100	10	on	on	ADP
ap-948	100	11	the	the	DET
ap-948	100	12	training	training	NOUN
ap-948	100	13	set	set	VERB
ap-948	100	14	only	only	ADV
ap-948	100	15	.	.	PUNCT
ap-948	101	1	this	this	PRON
ap-948	101	2	is	be	AUX
ap-948	101	3	necessary	necessary	ADJ
ap-948	101	4	for	for	ADP
ap-948	101	5	a	a	DET
ap-948	101	6	cross	cross	NOUN
ap-948	101	7	evaluation	evaluation	NOUN
ap-948	101	8	study	study	NOUN
ap-948	101	9	that	that	PRON
ap-948	101	10	requires	require	VERB
ap-948	101	11	that	that	SCONJ
ap-948	101	12	tests	test	NOUN
ap-948	101	13	are	be	AUX
ap-948	101	14	only	only	ADV
ap-948	101	15	performed	perform	VERB
ap-948	101	16	on	on	ADP
ap-948	101	17	data	datum	NOUN
ap-948	101	18	that	that	PRON
ap-948	101	19	has	have	AUX
ap-948	101	20	not	not	PART
ap-948	101	21	been	be	AUX
ap-948	101	22	part	part	NOUN
ap-948	101	23	of	of	ADP
ap-948	101	24	the	the	DET
ap-948	101	25	training	training	NOUN
ap-948	101	26	process	process	NOUN
ap-948	101	27	.	.	PUNCT
ap-948	102	1	this	this	DET
ap-948	102	2	dataset	dataset	NOUN
ap-948	102	3	is	be	AUX
ap-948	102	4	split	split	VERB
ap-948	102	5	up	up	ADP
ap-948	102	6	into	into	ADP
ap-948	102	7	46	46	NUM
ap-948	102	8	subsets	subset	NOUN
ap-948	102	9	containing	contain	VERB
ap-948	102	10	approx	approx	PROPN
ap-948	102	11	.	.	PUNCT
ap-948	103	1	49	49	NUM
ap-948	103	2	images	image	NOUN
ap-948	103	3	each	each	PRON
ap-948	103	4	.	.	PUNCT
ap-948	104	1	during	during	ADP
ap-948	104	2	cross	cross	NOUN
ap-948	104	3	evaluation	evaluation	NOUN
ap-948	104	4	,	,	PUNCT
ap-948	104	5	a	a	DET
ap-948	104	6	training	training	NOUN
ap-948	104	7	corpus	corpus	NOUN
ap-948	104	8	is	be	AUX
ap-948	104	9	formed	form	VERB
ap-948	104	10	of	of	ADP
ap-948	104	11	a	a	DET
ap-948	104	12	random	random	ADJ
ap-948	104	13	selection	selection	NOUN
ap-948	104	14	of	of	ADP
ap-948	104	15	45	45	NUM
ap-948	104	16	subsets	subset	NOUN
ap-948	104	17	,	,	PUNCT
ap-948	104	18	leaving	leave	VERB
ap-948	104	19	a	a	DET
ap-948	104	20	single	single	ADJ
ap-948	104	21	subset	subset	NOUN
ap-948	104	22	as	as	ADP
ap-948	104	23	a	a	DET
ap-948	104	24	test	test	NOUN
ap-948	104	25	set	set	NOUN
ap-948	104	26	.	.	PUNCT
ap-948	105	1	the	the	DET
ap-948	105	2	results	result	NOUN
ap-948	105	3	of	of	ADP
ap-948	105	4	such	such	ADJ
ap-948	105	5	evaluation	evaluation	NOUN
ap-948	105	6	runs	run	NOUN
ap-948	105	7	are	be	AUX
ap-948	105	8	averaged	average	VERB
ap-948	105	9	for	for	ADP
ap-948	105	10	the	the	DET
ap-948	105	11	final	final	ADJ
ap-948	105	12	results	result	NOUN
ap-948	105	13	presented	present	VERB
ap-948	105	14	in	in	ADP
ap-948	105	15	section	section	NOUN
ap-948	105	16	5	5	NUM
ap-948	105	17	.	.	PUNCT
ap-948	106	1	a	a	DET
ap-948	106	2	single	single	ADJ
ap-948	106	3	run	run	NOUN
ap-948	106	4	consists	consist	VERB
ap-948	106	5	of	of	ADP
ap-948	106	6	the	the	DET
ap-948	106	7	following	follow	VERB
ap-948	106	8	steps	step	NOUN
ap-948	106	9	:	:	PUNCT
ap-948	107	1	1	1	X
ap-948	107	2	.	.	X
ap-948	108	1	all	all	DET
ap-948	108	2	sift	sift	ADJ
ap-948	108	3	descriptors	descriptor	NOUN
ap-948	108	4	of	of	ADP
ap-948	108	5	the	the	DET
ap-948	108	6	test	test	NOUN
ap-948	108	7	set	set	NOUN
ap-948	108	8	are	be	AUX
ap-948	108	9	projected	project	VERB
ap-948	108	10	into	into	ADP
ap-948	108	11	the	the	DET
ap-948	108	12	feature	feature	NOUN
ap-948	108	13	space	space	NOUN
ap-948	108	14	found	find	VERB
ap-948	108	15	in	in	ADP
ap-948	108	16	3.1	3.1	NUM
ap-948	108	17	.	.	NOUN
ap-948	109	1	2	2	NUM
ap-948	109	2	.	.	X
ap-948	109	3	for	for	ADP
ap-948	109	4	all	all	DET
ap-948	109	5	such	such	ADJ
ap-948	109	6	features	feature	NOUN
ap-948	109	7	the	the	DET
ap-948	109	8	respective	respective	ADJ
ap-948	109	9	nearest	near	ADJ
ap-948	109	10	neighbors	neighbor	NOUN
ap-948	109	11	of	of	ADP
ap-948	109	12	the	the	DET
ap-948	109	13	target	target	NOUN
ap-948	109	14	clusters	cluster	NOUN
ap-948	109	15	from	from	ADP
ap-948	109	16	3.3	3.3	NUM
ap-948	109	17	is	be	AUX
ap-948	109	18	calculated	calculate	VERB
ap-948	109	19	.	.	PUNCT
ap-948	110	1	3	3	X
ap-948	110	2	.	.	X
ap-948	110	3	all	all	PRON
ap-948	110	4	features	feature	VERB
ap-948	110	5	whose	whose	DET
ap-948	110	6	distance	distance	NOUN
ap-948	110	7	to	to	ADP
ap-948	110	8	their	their	PRON
ap-948	110	9	target	target	NOUN
ap-948	110	10	cluster	cluster	NOUN
ap-948	110	11	is	be	AUX
ap-948	110	12	smaller	small	ADJ
ap-948	110	13	than	than	ADP
ap-948	110	14	a	a	DET
ap-948	110	15	given	give	VERB
ap-948	110	16	threshold	threshold	NOUN
ap-948	110	17	,	,	PUNCT
ap-948	110	18	are	be	AUX
ap-948	110	19	classified	classify	VERB
ap-948	110	20	as	as	ADP
ap-948	110	21	facial	facial	ADJ
ap-948	110	22	features	feature	NOUN
ap-948	110	23	.	.	PUNCT
ap-948	111	1	4	4	X
ap-948	111	2	.	.	X
ap-948	111	3	from	from	ADP
ap-948	111	4	the	the	DET
ap-948	111	5	annotation	annotation	NOUN
ap-948	111	6	of	of	ADP
ap-948	111	7	the	the	DET
ap-948	111	8	dataset	dataset	NOUN
ap-948	111	9	,	,	PUNCT
ap-948	111	10	true	true	ADJ
ap-948	111	11	positives	positive	NOUN
ap-948	111	12	and	and	CCONJ
ap-948	111	13	false	false	ADJ
ap-948	111	14	positives	positive	NOUN
ap-948	111	15	are	be	AUX
ap-948	111	16	discriminated	discriminate	VERB
ap-948	111	17	.	.	PUNCT
ap-948	112	1	5	5	X
ap-948	112	2	.	.	PUNCT
ap-948	112	3	by	by	ADP
ap-948	112	4	varying	vary	VERB
ap-948	112	5	the	the	DET
ap-948	112	6	threshold	threshold	NOUN
ap-948	112	7	,	,	PUNCT
ap-948	112	8	classifier	classifier	NOUN
ap-948	112	9	sensitivity	sensitivity	NOUN
ap-948	112	10	is	be	AUX
ap-948	112	11	adjusted	adjust	VERB
ap-948	112	12	.	.	PUNCT
ap-948	113	1	a	a	DET
ap-948	113	2	roc	roc	PROPN
ap-948	113	3	curve	curve	NOUN
ap-948	113	4	over	over	ADP
ap-948	113	5	true	true	ADJ
ap-948	113	6	positives	positive	NOUN
ap-948	113	7	vs.	vs.	ADP
ap-948	113	8	false	false	ADJ
ap-948	113	9	positives	positive	NOUN
ap-948	113	10	results	result	NOUN
ap-948	113	11	.	.	PUNCT
ap-948	114	1	this	this	DET
ap-948	114	2	procedure	procedure	NOUN
ap-948	114	3	is	be	AUX
ap-948	114	4	repeated	repeat	VERB
ap-948	114	5	for	for	ADP
ap-948	114	6	a	a	DET
ap-948	114	7	number	number	NOUN
ap-948	114	8	of	of	ADP
ap-948	114	9	cluster	cluster	NOUN
ap-948	114	10	densities	density	NOUN
ap-948	114	11	(	(	PUNCT
ap-948	114	12	see	see	VERB
ap-948	114	13	fig	fig	NOUN
ap-948	114	14	.	.	PUNCT
ap-948	115	1	5	5	NUM
ap-948	115	2	)	)	PUNCT
ap-948	115	3	to	to	PART
ap-948	115	4	find	find	VERB
ap-948	115	5	the	the	DET
ap-948	115	6	ideal	ideal	ADJ
ap-948	115	7	number	number	NOUN
ap-948	115	8	of	of	ADP
ap-948	115	9	clusters	cluster	NOUN
ap-948	115	10	.	.	PUNCT
ap-948	116	1	5	5	NUM
ap-948	116	2	results	result	NOUN
ap-948	116	3	in	in	ADP
ap-948	116	4	this	this	DET
ap-948	116	5	section	section	NOUN
ap-948	116	6	the	the	DET
ap-948	116	7	results	result	NOUN
ap-948	116	8	of	of	ADP
ap-948	116	9	the	the	DET
ap-948	116	10	previously	previously	ADV
ap-948	116	11	introduced	introduce	VERB
ap-948	116	12	interest	interest	NOUN
ap-948	116	13	point	point	NOUN
ap-948	116	14	detector	detector	NOUN
ap-948	116	15	and	and	CCONJ
ap-948	116	16	classifier	classifier	NOUN
ap-948	116	17	are	be	AUX
ap-948	116	18	presented	present	VERB
ap-948	116	19	.	.	PUNCT
ap-948	117	1	our	our	PRON
ap-948	117	2	dataset	dataset	NOUN
ap-948	117	3	consists	consist	NOUN
ap-948	117	4	of	of	ADP
ap-948	117	5	2254	2254	NUM
ap-948	117	6	images	image	NOUN
ap-948	117	7	showing	show	VERB
ap-948	117	8	one	one	NUM
ap-948	117	9	to	to	ADP
ap-948	117	10	three	three	NUM
ap-948	117	11	people	people	NOUN
ap-948	117	12	in	in	ADP
ap-948	117	13	cluttered	cluttered	ADJ
ap-948	117	14	backgrounds	background	NOUN
ap-948	117	15	.	.	PUNCT
ap-948	118	1	as	as	SCONJ
ap-948	118	2	described	describe	VERB
ap-948	118	3	in	in	ADP
ap-948	118	4	section	section	NOUN
ap-948	118	5	4	4	NUM
ap-948	118	6	the	the	DET
ap-948	118	7	dataset	dataset	NOUN
ap-948	118	8	is	be	AUX
ap-948	118	9	randomly	randomly	ADV
ap-948	118	10	split	split	VERB
ap-948	118	11	into	into	ADP
ap-948	118	12	a	a	DET
ap-948	118	13	training	training	NOUN
ap-948	118	14	set	set	NOUN
ap-948	118	15	containing	contain	VERB
ap-948	118	16	2205	2205	NUM
ap-948	118	17	images	image	NOUN
ap-948	118	18	and	and	CCONJ
ap-948	118	19	a	a	DET
ap-948	118	20	test	test	NOUN
ap-948	118	21	set	set	NOUN
ap-948	118	22	containing	contain	VERB
ap-948	118	23	49	49	NUM
ap-948	118	24	images	image	NOUN
ap-948	118	25	.	.	PUNCT
ap-948	119	1	fig	fig	NOUN
ap-948	119	2	.	.	PUNCT
ap-948	120	1	4	4	NUM
ap-948	120	2	depicts	depict	VERB
ap-948	120	3	the	the	DET
ap-948	120	4	distribution	distribution	NOUN
ap-948	120	5	of	of	ADP
ap-948	120	6	the	the	DET
ap-948	120	7	interest	interest	NOUN
ap-948	120	8	points	point	NOUN
ap-948	120	9	over	over	ADP
ap-948	120	10	the	the	DET
ap-948	120	11	face	face	NOUN
ap-948	120	12	area	area	NOUN
ap-948	120	13	.	.	PUNCT
ap-948	121	1	about	about	ADV
ap-948	121	2	150,000	150,000	NUM
ap-948	121	3	facial	facial	ADJ
ap-948	121	4	interest	interest	NOUN
ap-948	121	5	points	point	NOUN
ap-948	121	6	have	have	AUX
ap-948	121	7	been	be	AUX
ap-948	121	8	extracted	extract	VERB
ap-948	121	9	from	from	ADP
ap-948	121	10	all	all	DET
ap-948	121	11	images	image	NOUN
ap-948	121	12	of	of	ADP
ap-948	121	13	the	the	DET
ap-948	121	14	whole	whole	ADJ
ap-948	121	15	dataset	dataset	NOUN
ap-948	121	16	.	.	PUNCT
ap-948	122	1	they	they	PRON
ap-948	122	2	are	be	AUX
ap-948	122	3	mapped	map	VERB
ap-948	122	4	into	into	ADP
ap-948	122	5	a	a	DET
ap-948	122	6	normalized	normalize	VERB
ap-948	122	7	coordinate	coordinate	NOUN
ap-948	122	8	system	system	NOUN
ap-948	122	9	to	to	PART
ap-948	122	10	achieve	achieve	VERB
ap-948	122	11	comparability	comparability	NOUN
ap-948	122	12	.	.	PUNCT
ap-948	123	1	as	as	SCONJ
ap-948	123	2	can	can	AUX
ap-948	123	3	be	be	AUX
ap-948	123	4	seen	see	VERB
ap-948	123	5	,	,	PUNCT
ap-948	123	6	eyes	eye	NOUN
ap-948	123	7	,	,	PUNCT
ap-948	123	8	nostrils	nostril	NOUN
ap-948	123	9	and	and	CCONJ
ap-948	123	10	corners	corner	NOUN
ap-948	123	11	of	of	ADP
ap-948	123	12	the	the	DET
ap-948	123	13	mouth	mouth	NOUN
ap-948	123	14	are	be	AUX
ap-948	123	15	extracted	extract	VERB
ap-948	123	16	very	very	ADV
ap-948	123	17	reliably	reliably	ADV
ap-948	123	18	.	.	PUNCT
ap-948	124	1	thus	thus	ADV
ap-948	124	2	these	these	DET
ap-948	124	3	interest	interest	NOUN
ap-948	124	4	points	point	NOUN
ap-948	124	5	are	be	AUX
ap-948	124	6	suitable	suitable	ADJ
ap-948	124	7	for	for	ADP
ap-948	124	8	facial	facial	ADJ
ap-948	124	9	analysis	analysis	NOUN
ap-948	124	10	.	.	PUNCT
ap-948	125	1	in	in	ADP
ap-948	125	2	order	order	NOUN
ap-948	125	3	to	to	PART
ap-948	125	4	determine	determine	VERB
ap-948	125	5	the	the	DET
ap-948	125	6	best	good	ADJ
ap-948	125	7	setup	setup	NOUN
ap-948	125	8	of	of	ADP
ap-948	125	9	the	the	DET
ap-948	125	10	classifier	classifier	NOUN
ap-948	125	11	,	,	PUNCT
ap-948	125	12	the	the	DET
ap-948	125	13	influence	influence	NOUN
ap-948	125	14	of	of	ADP
ap-948	125	15	the	the	DET
ap-948	125	16	number	number	NOUN
ap-948	125	17	of	of	ADP
ap-948	125	18	clusters	cluster	NOUN
ap-948	125	19	on	on	ADP
ap-948	125	20	its	its	PRON
ap-948	125	21	performance	performance	NOUN
ap-948	125	22	has	have	AUX
ap-948	125	23	been	be	AUX
ap-948	125	24	analyzed	analyze	VERB
ap-948	125	25	.	.	PUNCT
ap-948	126	1	the	the	DET
ap-948	126	2	result	result	NOUN
ap-948	126	3	is	be	AUX
ap-948	126	4	shown	show	VERB
ap-948	126	5	in	in	ADP
ap-948	126	6	fig	fig	NOUN
ap-948	126	7	.	.	PUNCT
ap-948	127	1	5	5	X
ap-948	127	2	.	.	X
ap-948	128	1	obviously	obviously	ADV
ap-948	128	2	,	,	PUNCT
ap-948	128	3	the	the	DET
ap-948	128	4	effect	effect	NOUN
ap-948	128	5	of	of	ADP
ap-948	128	6	overfitting	overfitting	NOUN
ap-948	128	7	arises	arise	VERB
ap-948	128	8	for	for	ADP
ap-948	128	9	a	a	DET
ap-948	128	10	large	large	ADJ
ap-948	128	11	number	number	NOUN
ap-948	128	12	of	of	ADP
ap-948	128	13	clusters	cluster	NOUN
ap-948	128	14	and	and	CCONJ
ap-948	128	15	a	a	DET
ap-948	128	16	low	low	ADJ
ap-948	128	17	cluster	cluster	NOUN
ap-948	128	18	density	density	NOUN
ap-948	128	19	,	,	PUNCT
ap-948	128	20	respectively	respectively	ADV
ap-948	128	21	.	.	PUNCT
ap-948	129	1	fig	fig	NOUN
ap-948	129	2	.	.	PUNCT
ap-948	130	1	6	6	NUM
ap-948	130	2	shows	show	VERB
ap-948	130	3	the	the	DET
ap-948	130	4	distances	distance	NOUN
ap-948	130	5	of	of	ADP
ap-948	130	6	the	the	DET
ap-948	130	7	target	target	NOUN
ap-948	130	8	clusters	cluster	NOUN
ap-948	130	9	to	to	ADP
ap-948	130	10	the	the	DET
ap-948	130	11	features	feature	NOUN
ap-948	130	12	f	f	PROPN
ap-948	130	13	extracted	extract	VERB
ap-948	130	14	from	from	ADP
ap-948	130	15	the	the	DET
ap-948	130	16	test	test	NOUN
ap-948	130	17	set	set	NOUN
ap-948	130	18	.	.	PUNCT
ap-948	131	1	just	just	ADV
ap-948	131	2	as	as	SCONJ
ap-948	131	3	has	have	AUX
ap-948	131	4	been	be	AUX
ap-948	131	5	observed	observe	VERB
ap-948	131	6	from	from	ADP
ap-948	131	7	the	the	DET
ap-948	131	8	training	training	NOUN
ap-948	131	9	set	set	NOUN
ap-948	131	10	,	,	PUNCT
ap-948	131	11	the	the	DET
ap-948	131	12	test	test	NOUN
ap-948	131	13	features	feature	NOUN
ap-948	131	14	assigned	assign	VERB
ap-948	131	15	to	to	ADP
ap-948	131	16	a	a	DET
ap-948	131	17	face	face	NOUN
ap-948	131	18	(	(	PUNCT
ap-948	131	19	a	a	DET
ap-948	131	20	�	�	PROPN
ap-948	131	21	1	1	NUM
ap-948	131	22	)	)	PUNCT
ap-948	131	23	are	be	AUX
ap-948	131	24	closer	close	ADJ
ap-948	131	25	to	to	ADP
ap-948	131	26	the	the	DET
ap-948	131	27	target	target	NOUN
ap-948	131	28	cluster	cluster	NOUN
ap-948	131	29	than	than	ADP
ap-948	131	30	the	the	DET
ap-948	131	31	other	other	ADJ
ap-948	131	32	features	feature	NOUN
ap-948	131	33	(	(	PUNCT
ap-948	131	34	a	a	DET
ap-948	131	35	�	�	PROPN
ap-948	131	36	0	0	NUM
ap-948	131	37	)	)	PUNCT
ap-948	131	38	.	.	PUNCT
ap-948	132	1	since	since	SCONJ
ap-948	132	2	only	only	ADV
ap-948	132	3	a	a	DET
ap-948	132	4	small	small	ADJ
ap-948	132	5	part	part	NOUN
ap-948	132	6	of	of	ADP
ap-948	132	7	the	the	DET
ap-948	132	8	image	image	NOUN
ap-948	132	9	area	area	NOUN
ap-948	132	10	is	be	AUX
ap-948	132	11	covered	cover	VERB
ap-948	132	12	by	by	ADP
ap-948	132	13	human	human	ADJ
ap-948	132	14	faces	face	NOUN
ap-948	132	15	,	,	PUNCT
ap-948	132	16	the	the	DET
ap-948	132	17	number	number	NOUN
ap-948	132	18	of	of	ADP
ap-948	132	19	interest	interest	NOUN
ap-948	132	20	points	point	NOUN
ap-948	132	21	describing	describe	VERB
ap-948	132	22	facial	facial	ADJ
ap-948	132	23	features	feature	NOUN
ap-948	132	24	is	be	AUX
ap-948	132	25	20	20	NUM
ap-948	132	26	times	time	NOUN
ap-948	132	27	lower	low	ADJ
ap-948	132	28	than	than	ADP
ap-948	132	29	the	the	DET
ap-948	132	30	number	number	NOUN
ap-948	132	31	of	of	ADP
ap-948	132	32	interest	interest	NOUN
ap-948	132	33	points	point	NOUN
ap-948	132	34	found	find	VERB
ap-948	132	35	in	in	ADP
ap-948	132	36	the	the	DET
ap-948	132	37	remaining	remain	VERB
ap-948	132	38	image	image	NOUN
ap-948	132	39	.	.	PUNCT
ap-948	133	1	that	that	PRON
ap-948	133	2	is	be	AUX
ap-948	133	3	why	why	SCONJ
ap-948	133	4	only	only	ADV
ap-948	133	5	really	really	ADV
ap-948	133	6	low	low	ADJ
ap-948	133	7	false	false	ADJ
ap-948	133	8	-	-	PUNCT
ap-948	133	9	positive	positive	ADJ
ap-948	133	10	rates	rate	NOUN
ap-948	133	11	(	(	PUNCT
ap-948	133	12	0.1	0.1	NUM
ap-948	133	13	%	%	NOUN
ap-948	133	14	–	–	PUNCT
ap-948	133	15	0.2	0.2	NUM
ap-948	133	16	%	%	NOUN
ap-948	133	17	)	)	PUNCT
ap-948	133	18	are	be	AUX
ap-948	133	19	acceptable	acceptable	ADJ
ap-948	133	20	.	.	PUNCT
ap-948	134	1	setting	set	VERB
ap-948	134	2	a	a	DET
ap-948	134	3	threshold	threshold	NOUN
ap-948	134	4	for	for	ADP
ap-948	134	5	0.2	0.2	NUM
ap-948	134	6	%	%	NOUN
ap-948	134	7	false	false	ADJ
ap-948	134	8	positives	positive	NOUN
ap-948	134	9	results	result	NOUN
ap-948	134	10	in	in	ADP
ap-948	134	11	a	a	DET
ap-948	134	12	detection	detection	NOUN
ap-948	134	13	rate	rate	NOUN
ap-948	134	14	of	of	ADP
ap-948	134	15	about	about	ADV
ap-948	134	16	20	20	NUM
ap-948	134	17	%	%	NOUN
ap-948	134	18	,	,	PUNCT
ap-948	134	19	which	which	PRON
ap-948	134	20	seems	seem	VERB
ap-948	134	21	acceptable	acceptable	ADJ
ap-948	134	22	given	give	VERB
ap-948	134	23	the	the	DET
ap-948	134	24	fact	fact	NOUN
ap-948	134	25	that	that	SCONJ
ap-948	134	26	this	this	PRON
ap-948	134	27	corresponds	correspond	VERB
ap-948	134	28	to	to	ADP
ap-948	134	29	about	about	ADV
ap-948	134	30	10	10	NUM
ap-948	134	31	correctly	correctly	ADV
ap-948	134	32	detected	detect	VERB
ap-948	134	33	facial	facial	ADJ
ap-948	134	34	fea	fea	NOUN
ap-948	134	35	©	©	PROPN
ap-948	134	36	czech	czech	PROPN
ap-948	134	37	technical	technical	PROPN
ap-948	134	38	university	university	PROPN
ap-948	134	39	publishing	publishing	NOUN
ap-948	134	40	house	house	NOUN
ap-948	134	41	http://ctn.cvut.cz/ap/	http://ctn.cvut.cz/ap/	PROPN
ap-948	134	42	5	5	NUM
ap-948	134	43	acta	acta	PROPN
ap-948	134	44	polytechnica	polytechnica	PROPN
ap-948	134	45	vol	vol	NOUN
ap-948	134	46	.	.	PUNCT
ap-948	135	1	47	47	NUM
ap-948	135	2	no	no	NOUN
ap-948	135	3	.	.	PUNCT
ap-948	136	1	4–5/2007	4–5/2007	NUM
ap-948	136	2	fig	fig	NOUN
ap-948	136	3	.	.	PUNCT
ap-948	137	1	3	3	NUM
ap-948	137	2	:	:	PUNCT
ap-948	137	3	distances	distance	NOUN
ap-948	137	4	of	of	ADP
ap-948	137	5	the	the	DET
ap-948	137	6	training	training	NOUN
ap-948	137	7	feature	feature	NOUN
ap-948	137	8	vectors	vector	NOUN
ap-948	137	9	f	f	X
ap-948	137	10	to	to	ADP
ap-948	137	11	the	the	DET
ap-948	137	12	target	target	NOUN
ap-948	137	13	cluster	cluster	NOUN
ap-948	137	14	.	.	PUNCT
ap-948	138	1	distribution	distribution	NOUN
ap-948	138	2	of	of	ADP
ap-948	138	3	interest	interest	NOUN
ap-948	138	4	in	in	ADP
ap-948	138	5	a	a	DET
ap-948	138	6	normalized	normalize	VERB
ap-948	138	7	face	face	NOUN
ap-948	138	8	fig	fig	NOUN
ap-948	138	9	.	.	PUNCT
ap-948	139	1	4	4	NUM
ap-948	139	2	:	:	PUNCT
ap-948	139	3	this	this	DET
ap-948	139	4	figure	figure	NOUN
ap-948	139	5	shows	show	VERB
ap-948	139	6	the	the	DET
ap-948	139	7	distribution	distribution	NOUN
ap-948	139	8	of	of	ADP
ap-948	139	9	interest	interest	NOUN
ap-948	139	10	points	point	NOUN
ap-948	139	11	detected	detect	VERB
ap-948	139	12	in	in	ADP
ap-948	139	13	a	a	DET
ap-948	139	14	face	face	NOUN
ap-948	139	15	by	by	ADP
ap-948	139	16	the	the	DET
ap-948	139	17	scale	scale	NOUN
ap-948	139	18	adapted	adapt	VERB
ap-948	139	19	harris	harris	PROPN
ap-948	139	20	corner	corner	PROPN
ap-948	139	21	detector	detector	NOUN
ap-948	139	22	.	.	PUNCT
ap-948	140	1	the	the	DET
ap-948	140	2	interest	interest	NOUN
ap-948	140	3	points	point	NOUN
ap-948	140	4	are	be	AUX
ap-948	140	5	mapped	map	VERB
ap-948	140	6	into	into	ADP
ap-948	140	7	the	the	DET
ap-948	140	8	normalized	normalize	VERB
ap-948	140	9	coordinate	coordinate	NOUN
ap-948	140	10	system	system	NOUN
ap-948	140	11	.	.	PUNCT
ap-948	141	1	ture	ture	PROPN
ap-948	141	2	points	point	NOUN
ap-948	141	3	per	per	ADP
ap-948	141	4	image	image	NOUN
ap-948	141	5	,	,	PUNCT
ap-948	141	6	whereas	whereas	SCONJ
ap-948	141	7	approximately	approximately	ADV
ap-948	141	8	2	2	NUM
ap-948	141	9	features	feature	NOUN
ap-948	141	10	are	be	AUX
ap-948	141	11	wrongly	wrongly	ADV
ap-948	141	12	classified	classify	VERB
ap-948	141	13	.	.	PUNCT
ap-948	142	1	this	this	DET
ap-948	142	2	aspect	aspect	NOUN
ap-948	142	3	is	be	AUX
ap-948	142	4	exemplified	exemplify	VERB
ap-948	142	5	in	in	ADP
ap-948	142	6	the	the	DET
ap-948	142	7	following	following	NOUN
ap-948	142	8	.	.	PUNCT
ap-948	143	1	fig	fig	NOUN
ap-948	143	2	.	.	PUNCT
ap-948	144	1	7	7	NUM
ap-948	144	2	shows	show	VERB
ap-948	144	3	all	all	DET
ap-948	144	4	interest	interest	NOUN
ap-948	144	5	points	point	NOUN
ap-948	144	6	detected	detect	VERB
ap-948	144	7	by	by	ADP
ap-948	144	8	the	the	DET
ap-948	144	9	scale	scale	NOUN
ap-948	144	10	-	-	PUNCT
ap-948	144	11	adapted	adapt	VERB
ap-948	144	12	harris	harris	PROPN
ap-948	144	13	corner	corner	PROPN
ap-948	144	14	detector	detector	NOUN
ap-948	144	15	.	.	PUNCT
ap-948	145	1	these	these	DET
ap-948	145	2	points	point	NOUN
ap-948	145	3	,	,	PUNCT
ap-948	145	4	indicated	indicate	VERB
ap-948	145	5	by	by	ADP
ap-948	145	6	circles	circle	NOUN
ap-948	145	7	,	,	PUNCT
ap-948	145	8	are	be	AUX
ap-948	145	9	especially	especially	ADV
ap-948	145	10	located	locate	VERB
ap-948	145	11	in	in	ADP
ap-948	145	12	corner	corner	NOUN
ap-948	145	13	like	like	ADP
ap-948	145	14	structures	structure	NOUN
ap-948	145	15	.	.	PUNCT
ap-948	146	1	the	the	DET
ap-948	146	2	result	result	NOUN
ap-948	146	3	of	of	ADP
ap-948	146	4	the	the	DET
ap-948	146	5	classifier	classifier	NOUN
ap-948	146	6	can	can	AUX
ap-948	146	7	be	be	AUX
ap-948	146	8	seen	see	VERB
ap-948	146	9	in	in	ADP
ap-948	146	10	fig	fig	NOUN
ap-948	146	11	.	.	PUNCT
ap-948	147	1	8	8	X
ap-948	147	2	.	.	PUNCT
ap-948	148	1	it	it	PRON
ap-948	148	2	shows	show	VERB
ap-948	148	3	only	only	ADV
ap-948	148	4	the	the	DET
ap-948	148	5	interest	interest	NOUN
ap-948	148	6	points	point	NOUN
ap-948	148	7	that	that	PRON
ap-948	148	8	have	have	AUX
ap-948	148	9	been	be	AUX
ap-948	148	10	classified	classify	VERB
ap-948	148	11	as	as	ADP
ap-948	148	12	facial	facial	ADJ
ap-948	148	13	features	feature	NOUN
ap-948	148	14	.	.	PUNCT
ap-948	149	1	these	these	DET
ap-948	149	2	features	feature	NOUN
ap-948	149	3	represent	represent	VERB
ap-948	149	4	eyes	eye	NOUN
ap-948	149	5	,	,	PUNCT
ap-948	149	6	eye	eye	NOUN
ap-948	149	7	brows	brow	NOUN
ap-948	149	8	and	and	CCONJ
ap-948	149	9	corner	corner	NOUN
ap-948	149	10	of	of	ADP
ap-948	149	11	the	the	DET
ap-948	149	12	mouth	mouth	NOUN
ap-948	149	13	.	.	PUNCT
ap-948	150	1	6	6	NUM
ap-948	150	2	conclusion	conclusion	NOUN
ap-948	150	3	this	this	DET
ap-948	150	4	paper	paper	NOUN
ap-948	150	5	has	have	AUX
ap-948	150	6	introduced	introduce	VERB
ap-948	150	7	a	a	DET
ap-948	150	8	new	new	ADJ
ap-948	150	9	approach	approach	NOUN
ap-948	150	10	to	to	ADP
ap-948	150	11	the	the	DET
ap-948	150	12	detection	detection	NOUN
ap-948	150	13	of	of	ADP
ap-948	150	14	facial	facial	ADJ
ap-948	150	15	features	feature	NOUN
ap-948	150	16	.	.	PUNCT
ap-948	151	1	our	our	PRON
ap-948	151	2	approach	approach	NOUN
ap-948	151	3	builds	build	VERB
ap-948	151	4	on	on	ADP
ap-948	151	5	local	local	ADJ
ap-948	151	6	image	image	NOUN
ap-948	151	7	descriptions	description	NOUN
ap-948	151	8	that	that	PRON
ap-948	151	9	are	be	AUX
ap-948	151	10	invariant	invariant	ADJ
ap-948	151	11	towards	towards	ADP
ap-948	151	12	affine	affine	NOUN
ap-948	151	13	image	image	NOUN
ap-948	151	14	transformations	transformation	NOUN
ap-948	151	15	and	and	CCONJ
ap-948	151	16	illumination	illumination	NOUN
ap-948	151	17	.	.	PUNCT
ap-948	152	1	we	we	PRON
ap-948	152	2	have	have	AUX
ap-948	152	3	shown	show	VERB
ap-948	152	4	that	that	SCONJ
ap-948	152	5	the	the	DET
ap-948	152	6	scale	scale	NOUN
ap-948	152	7	adapted	adapt	VERB
ap-948	152	8	harris	harris	PROPN
ap-948	152	9	detector	detector	NOUN
ap-948	152	10	yields	yield	NOUN
ap-948	152	11	feature	feature	NOUN
ap-948	152	12	points	point	NOUN
ap-948	152	13	that	that	PRON
ap-948	152	14	are	be	AUX
ap-948	152	15	suitable	suitable	ADJ
ap-948	152	16	for	for	ADP
ap-948	152	17	the	the	DET
ap-948	152	18	detection	detection	NOUN
ap-948	152	19	of	of	ADP
ap-948	152	20	specific	specific	ADJ
ap-948	152	21	facial	facial	ADJ
ap-948	152	22	features	feature	NOUN
ap-948	152	23	like	like	ADP
ap-948	152	24	eyes	eye	NOUN
ap-948	152	25	,	,	PUNCT
ap-948	152	26	nostrils	nostril	NOUN
ap-948	152	27	and	and	CCONJ
ap-948	152	28	corners	corner	NOUN
ap-948	152	29	of	of	ADP
ap-948	152	30	the	the	DET
ap-948	152	31	mouth	mouth	NOUN
ap-948	152	32	,	,	PUNCT
ap-948	152	33	and	and	CCONJ
ap-948	152	34	that	that	SCONJ
ap-948	152	35	these	these	DET
ap-948	152	36	features	feature	NOUN
ap-948	152	37	can	can	AUX
ap-948	152	38	be	be	AUX
ap-948	152	39	appropriately	appropriately	ADV
ap-948	152	40	described	describe	VERB
ap-948	152	41	by	by	ADP
ap-948	152	42	the	the	DET
ap-948	152	43	sift	sift	PROPN
ap-948	152	44	descriptor	descriptor	NOUN
ap-948	152	45	.	.	PUNCT
ap-948	153	1	our	our	PRON
ap-948	153	2	new	new	ADJ
ap-948	153	3	method	method	NOUN
ap-948	153	4	is	be	AUX
ap-948	153	5	able	able	ADJ
ap-948	153	6	to	to	PART
ap-948	153	7	detect	detect	VERB
ap-948	153	8	on	on	ADP
ap-948	153	9	average	average	ADJ
ap-948	153	10	10	10	NUM
ap-948	153	11	features	feature	NOUN
ap-948	153	12	per	per	ADP
ap-948	153	13	face	face	NOUN
ap-948	153	14	,	,	PUNCT
ap-948	153	15	with	with	ADP
ap-948	153	16	a	a	DET
ap-948	153	17	false	false	ADJ
ap-948	153	18	-	-	PUNCT
ap-948	153	19	positive	positive	ADJ
ap-948	153	20	rate	rate	NOUN
ap-948	153	21	of	of	ADP
ap-948	153	22	0.2	0.2	NUM
ap-948	153	23	%	%	NOUN
ap-948	153	24	,	,	PUNCT
ap-948	153	25	which	which	PRON
ap-948	153	26	corresponds	correspond	VERB
ap-948	153	27	to	to	ADP
ap-948	153	28	approximately	approximately	ADV
ap-948	153	29	2	2	NUM
ap-948	153	30	wrongly	wrongly	ADV
ap-948	153	31	classified	classify	VERB
ap-948	153	32	features	feature	NOUN
ap-948	153	33	per	per	ADP
ap-948	153	34	image	image	NOUN
ap-948	153	35	.	.	PUNCT
ap-948	154	1	acknowledgements	acknowledgement	VERB
ap-948	154	2	the	the	DET
ap-948	154	3	research	research	NOUN
ap-948	154	4	described	describe	VERB
ap-948	154	5	in	in	ADP
ap-948	154	6	this	this	DET
ap-948	154	7	paper	paper	NOUN
ap-948	154	8	was	be	AUX
ap-948	154	9	supervised	supervise	VERB
ap-948	154	10	by	by	ADP
ap-948	154	11	prof	prof	PROPN
ap-948	154	12	.	.	PUNCT
ap-948	155	1	j.-r	j.-r	PROPN
ap-948	155	2	.	.	PUNCT
ap-948	156	1	ohm	ohm	PROPN
ap-948	156	2	,	,	PUNCT
ap-948	156	3	institute	institute	NOUN
ap-948	156	4	of	of	ADP
ap-948	156	5	communications	communication	NOUN
ap-948	156	6	engineering	engineering	NOUN
ap-948	156	7	at	at	ADP
ap-948	156	8	rwth	rwth	PROPN
ap-948	156	9	aachen	aachen	PROPN
ap-948	156	10	university	university	PROPN
ap-948	156	11	.	.	PUNCT
ap-948	157	1	references	reference	NOUN
ap-948	157	2	[	[	X
ap-948	157	3	1	1	NUM
ap-948	157	4	]	]	X
ap-948	157	5	bellman	bellman	NOUN
ap-948	157	6	,	,	PUNCT
ap-948	157	7	r.	r.	PROPN
ap-948	157	8	:	:	PUNCT
ap-948	157	9	adaptive	adaptive	PROPN
ap-948	157	10	control	control	NOUN
ap-948	157	11	processes	process	NOUN
ap-948	157	12	:	:	PUNCT
ap-948	157	13	a	a	DET
ap-948	157	14	guided	guide	VERB
ap-948	157	15	tour	tour	NOUN
ap-948	157	16	.	.	PUNCT
ap-948	158	1	princeton	princeton	PROPN
ap-948	158	2	university	university	PROPN
ap-948	158	3	press	press	NOUN
ap-948	158	4	,	,	PUNCT
ap-948	158	5	1961	1961	NUM
ap-948	158	6	.	.	PUNCT
ap-948	159	1	[	[	X
ap-948	159	2	2	2	NUM
ap-948	159	3	]	]	PUNCT
ap-948	159	4	fukunaga	fukunaga	PROPN
ap-948	159	5	,	,	PUNCT
ap-948	159	6	k.	k.	NOUN
ap-948	159	7	:	:	PUNCT
ap-948	159	8	introduction	introduction	NOUN
ap-948	159	9	in	in	ADP
ap-948	159	10	statistical	statistical	ADJ
ap-948	159	11	pattern	pattern	NOUN
ap-948	159	12	recognition	recognition	NOUN
ap-948	159	13	.	.	PUNCT
ap-948	160	1	1972	1972	NUM
ap-948	160	2	.	.	PUNCT
ap-948	161	1	[	[	X
ap-948	161	2	3	3	NUM
ap-948	161	3	]	]	X
ap-948	161	4	harris	harris	PROPN
ap-948	161	5	,	,	PUNCT
ap-948	161	6	c.	c.	PROPN
ap-948	161	7	,	,	PUNCT
ap-948	161	8	stephens	stephen	VERB
ap-948	161	9	:	:	PUNCT
ap-948	161	10	a	a	DET
ap-948	161	11	combined	combined	ADJ
ap-948	161	12	corner	corner	NOUN
ap-948	161	13	and	and	CCONJ
ap-948	161	14	edge	edge	NOUN
ap-948	161	15	detector	detector	NOUN
ap-948	161	16	.	.	PUNCT
ap-948	161	17	alvey	alvey	PROPN
ap-948	161	18	vision	vision	PROPN
ap-948	161	19	conference	conference	NOUN
ap-948	161	20	,	,	PUNCT
ap-948	161	21	1988	1988	NUM
ap-948	161	22	.	.	PUNCT
ap-948	162	1	[	[	X
ap-948	162	2	4	4	NUM
ap-948	162	3	]	]	PUNCT
ap-948	162	4	hjelmas	hjelma	NOUN
ap-948	162	5	,	,	PUNCT
ap-948	162	6	e.	e.	PROPN
ap-948	162	7	,	,	PUNCT
ap-948	162	8	low	low	PROPN
ap-948	162	9	,	,	PUNCT
ap-948	162	10	b.	b.	PROPN
ap-948	162	11	k.	k.	PROPN
ap-948	162	12	:	:	PUNCT
ap-948	162	13	face	face	VERB
ap-948	162	14	detection	detection	NOUN
ap-948	162	15	:	:	PUNCT
ap-948	162	16	a	a	DET
ap-948	162	17	survey	survey	NOUN
ap-948	162	18	.	.	PUNCT
ap-948	163	1	computer	computer	NOUN
ap-948	163	2	vision	vision	NOUN
ap-948	163	3	and	and	CCONJ
ap-948	163	4	image	image	NOUN
ap-948	163	5	understanding	understanding	NOUN
ap-948	163	6	,	,	PUNCT
ap-948	163	7	vol	vol	NOUN
ap-948	163	8	.	.	PROPN
ap-948	163	9	83	83	NUM
ap-948	163	10	(	(	PUNCT
ap-948	163	11	2001	2001	NUM
ap-948	163	12	)	)	PUNCT
ap-948	163	13	,	,	PUNCT
ap-948	163	14	no	no	INTJ
ap-948	163	15	.	.	NOUN
ap-948	163	16	3	3	NUM
ap-948	163	17	,	,	PUNCT
ap-948	163	18	p.	p.	NOUN
ap-948	163	19	236–274	236–274	NUM
ap-948	163	20	.	.	PUNCT
ap-948	164	1	[	[	X
ap-948	164	2	5	5	NUM
ap-948	164	3	]	]	X
ap-948	164	4	lindeberg	lindeberg	PROPN
ap-948	164	5	,	,	PUNCT
ap-948	164	6	t.	t.	NOUN
ap-948	164	7	:	:	PUNCT
ap-948	164	8	edge	edge	NOUN
ap-948	164	9	detection	detection	NOUN
ap-948	164	10	and	and	CCONJ
ap-948	164	11	ridge	ridge	NOUN
ap-948	164	12	detection	detection	NOUN
ap-948	164	13	with	with	ADP
ap-948	164	14	automatic	automatic	ADJ
ap-948	164	15	scale	scale	NOUN
ap-948	164	16	selection	selection	NOUN
ap-948	164	17	.	.	PUNCT
ap-948	165	1	international	international	ADJ
ap-948	165	2	journal	journal	PROPN
ap-948	165	3	of	of	ADP
ap-948	165	4	computer	computer	NOUN
ap-948	165	5	vision	vision	NOUN
ap-948	165	6	,	,	PUNCT
ap-948	165	7	1998	1998	NUM
ap-948	165	8	.	.	PUNCT
ap-948	166	1	[	[	X
ap-948	166	2	6	6	NUM
ap-948	166	3	]	]	X
ap-948	166	4	lowe	lowe	PROPN
ap-948	166	5	,	,	PUNCT
ap-948	166	6	d.	d.	PROPN
ap-948	166	7	g.	g.	PROPN
ap-948	166	8	:	:	PUNCT
ap-948	166	9	distinctive	distinctive	ADJ
ap-948	166	10	image	image	NOUN
ap-948	166	11	features	feature	NOUN
ap-948	166	12	from	from	ADP
ap-948	166	13	scale	scale	NOUN
ap-948	166	14	-	-	PUNCT
ap-948	166	15	invariant	invariant	ADJ
ap-948	166	16	keypoints	keypoint	NOUN
ap-948	166	17	.	.	PUNCT
ap-948	167	1	international	international	ADJ
ap-948	167	2	journal	journal	NOUN
ap-948	167	3	of	of	ADP
ap-948	167	4	computer	computer	NOUN
ap-948	167	5	vision	vision	NOUN
ap-948	167	6	,	,	PUNCT
ap-948	167	7	2004	2004	NUM
ap-948	167	8	.	.	PUNCT
ap-948	168	1	6	6	NUM
ap-948	168	2	©	©	PROPN
ap-948	168	3	czech	czech	PROPN
ap-948	168	4	technical	technical	PROPN
ap-948	168	5	university	university	PROPN
ap-948	168	6	publishing	publishing	NOUN
ap-948	168	7	house	house	NOUN
ap-948	168	8	http://ctn.cvut.cz/ap/	http://ctn.cvut.cz/ap/	PROPN
ap-948	168	9	acta	acta	PROPN
ap-948	168	10	polytechnica	polytechnica	PROPN
ap-948	168	11	vol	vol	NOUN
ap-948	168	12	.	.	PUNCT
ap-948	169	1	47	47	NUM
ap-948	169	2	no	no	NOUN
ap-948	169	3	.	.	PUNCT
ap-948	170	1	4–5/2007	4–5/2007	NUM
ap-948	170	2	fig	fig	NOUN
ap-948	170	3	.	.	PUNCT
ap-948	171	1	5	5	NUM
ap-948	171	2	:	:	PUNCT
ap-948	171	3	this	this	DET
ap-948	171	4	figure	figure	NOUN
ap-948	171	5	shows	show	VERB
ap-948	171	6	the	the	DET
ap-948	171	7	influence	influence	NOUN
ap-948	171	8	of	of	ADP
ap-948	171	9	different	different	ADJ
ap-948	171	10	cluster	cluster	NOUN
ap-948	171	11	densities	density	NOUN
ap-948	171	12	on	on	ADP
ap-948	171	13	the	the	DET
ap-948	171	14	performance	performance	NOUN
ap-948	171	15	of	of	ADP
ap-948	171	16	the	the	DET
ap-948	171	17	classifier	classifier	NOUN
ap-948	171	18	.	.	PUNCT
ap-948	172	1	thus	thus	ADV
ap-948	172	2	the	the	DET
ap-948	172	3	cluster	cluster	NOUN
ap-948	172	4	density	density	NOUN
ap-948	172	5	is	be	AUX
ap-948	172	6	a	a	DET
ap-948	172	7	measure	measure	NOUN
ap-948	172	8	for	for	ADP
ap-948	172	9	the	the	DET
ap-948	172	10	number	number	NOUN
ap-948	172	11	of	of	ADP
ap-948	172	12	training	training	NOUN
ap-948	172	13	features	feature	NOUN
ap-948	172	14	represented	represent	VERB
ap-948	172	15	by	by	ADP
ap-948	172	16	a	a	DET
ap-948	172	17	cluster	cluster	NOUN
ap-948	172	18	.	.	PUNCT
ap-948	173	1	the	the	DET
ap-948	173	2	smaller	small	ADJ
ap-948	173	3	the	the	DET
ap-948	173	4	value	value	NOUN
ap-948	173	5	,	,	PUNCT
ap-948	173	6	the	the	DET
ap-948	173	7	more	more	ADJ
ap-948	173	8	clusters	cluster	NOUN
ap-948	173	9	are	be	AUX
ap-948	173	10	computed	compute	VERB
ap-948	173	11	.	.	PUNCT
ap-948	174	1	fig	fig	NOUN
ap-948	174	2	.	.	PUNCT
ap-948	175	1	6	6	NUM
ap-948	175	2	:	:	PUNCT
ap-948	175	3	distances	distance	NOUN
ap-948	175	4	of	of	ADP
ap-948	175	5	the	the	DET
ap-948	175	6	test	test	NOUN
ap-948	175	7	feature	feature	NOUN
ap-948	175	8	vectors	vector	NOUN
ap-948	175	9	f	f	X
ap-948	175	10	to	to	ADP
ap-948	175	11	the	the	DET
ap-948	175	12	target	target	NOUN
ap-948	175	13	cluster	cluster	NOUN
ap-948	175	14	fig	fig	NOUN
ap-948	175	15	.	.	PUNCT
ap-948	176	1	7	7	NUM
ap-948	176	2	:	:	PUNCT
ap-948	176	3	this	this	DET
ap-948	176	4	image	image	NOUN
ap-948	176	5	of	of	ADP
ap-948	176	6	the	the	DET
ap-948	176	7	test	test	NOUN
ap-948	176	8	set	set	NOUN
ap-948	176	9	shows	show	VERB
ap-948	176	10	all	all	DET
ap-948	176	11	interest	interest	NOUN
ap-948	176	12	points	point	NOUN
ap-948	176	13	detected	detect	VERB
ap-948	176	14	by	by	ADP
ap-948	176	15	the	the	DET
ap-948	176	16	scale	scale	NOUN
ap-948	176	17	-	-	PUNCT
ap-948	176	18	adapted	adapt	VERB
ap-948	176	19	harris	harris	PROPN
ap-948	176	20	corner	corner	PROPN
ap-948	176	21	detector	detector	NOUN
ap-948	176	22	.	.	PUNCT
ap-948	177	1	the	the	DET
ap-948	177	2	diameter	diameter	NOUN
ap-948	177	3	of	of	ADP
ap-948	177	4	the	the	DET
ap-948	177	5	circle	circle	NOUN
ap-948	177	6	indicates	indicate	VERB
ap-948	177	7	the	the	DET
ap-948	177	8	appropriate	appropriate	ADJ
ap-948	177	9	scale	scale	NOUN
ap-948	177	10	on	on	ADP
ap-948	177	11	which	which	PRON
ap-948	177	12	the	the	DET
ap-948	177	13	interest	interest	NOUN
ap-948	177	14	point	point	NOUN
ap-948	177	15	has	have	AUX
ap-948	177	16	been	be	AUX
ap-948	177	17	detected	detect	VERB
ap-948	177	18	.	.	PUNCT
ap-948	178	1	fig	fig	NOUN
ap-948	178	2	.	.	PUNCT
ap-948	179	1	8	8	NUM
ap-948	179	2	:	:	PUNCT
ap-948	179	3	this	this	DET
ap-948	179	4	image	image	NOUN
ap-948	179	5	shows	show	VERB
ap-948	179	6	only	only	ADV
ap-948	179	7	the	the	DET
ap-948	179	8	interest	interest	NOUN
ap-948	179	9	points	point	NOUN
ap-948	179	10	that	that	PRON
ap-948	179	11	have	have	AUX
ap-948	179	12	been	be	AUX
ap-948	179	13	classified	classify	VERB
ap-948	179	14	as	as	ADP
ap-948	179	15	facial	facial	ADJ
ap-948	179	16	features	feature	NOUN
ap-948	179	17	[	[	X
ap-948	179	18	7	7	X
ap-948	179	19	]	]	PUNCT
ap-948	179	20	mikolajczy	mikolajczy	PROPN
ap-948	179	21	,	,	PUNCT
ap-948	179	22	k.	k.	PROPN
ap-948	179	23	,	,	PUNCT
ap-948	179	24	schmid	schmid	PROPN
ap-948	179	25	,	,	PUNCT
ap-948	179	26	c.	c.	PROPN
ap-948	179	27	:	:	PUNCT
ap-948	179	28	scale	scale	NOUN
ap-948	179	29	&	&	CCONJ
ap-948	179	30	affine	affine	PROPN
ap-948	179	31	invariant	invariant	ADJ
ap-948	179	32	interest	interest	NOUN
ap-948	179	33	point	point	NOUN
ap-948	179	34	detectors	detector	NOUN
ap-948	179	35	.	.	PUNCT
ap-948	180	1	international	international	ADJ
ap-948	180	2	journal	journal	NOUN
ap-948	180	3	of	of	ADP
ap-948	180	4	computer	computer	NOUN
ap-948	180	5	vision	vision	NOUN
ap-948	180	6	,	,	PUNCT
ap-948	180	7	2004	2004	NUM
ap-948	180	8	.	.	PUNCT
ap-948	181	1	[	[	X
ap-948	181	2	8	8	NUM
ap-948	181	3	]	]	X
ap-948	181	4	mikolajczy	mikolajczy	PROPN
ap-948	181	5	,	,	PUNCT
ap-948	181	6	k.	k.	PROPN
ap-948	181	7	,	,	PUNCT
ap-948	181	8	schmid	schmid	PROPN
ap-948	181	9	,	,	PUNCT
ap-948	181	10	c.	c.	PROPN
ap-948	181	11	:	:	PUNCT
ap-948	181	12	a	a	DET
ap-948	181	13	performance	performance	NOUN
ap-948	181	14	evaluation	evaluation	NOUN
ap-948	181	15	of	of	ADP
ap-948	181	16	local	local	ADJ
ap-948	181	17	descriptors	descriptor	NOUN
ap-948	181	18	.	.	PUNCT
ap-948	182	1	ieee	ieee	NOUN
ap-948	182	2	transactions	transaction	NOUN
ap-948	182	3	on	on	ADP
ap-948	182	4	pattern	pattern	NOUN
ap-948	182	5	analysis	analysis	NOUN
ap-948	182	6	and	and	CCONJ
ap-948	182	7	machine	machine	NOUN
ap-948	182	8	intelligence	intelligence	NOUN
ap-948	182	9	,	,	PUNCT
ap-948	182	10	2005	2005	NUM
ap-948	182	11	.	.	PUNCT
ap-948	183	1	[	[	X
ap-948	183	2	9	9	NUM
ap-948	183	3	]	]	PUNCT
ap-948	183	4	mikolajczyk	mikolajczyk	NOUN
ap-948	183	5	,	,	PUNCT
ap-948	183	6	k.	k.	PROPN
ap-948	183	7	,	,	PUNCT
ap-948	183	8	leibe	leibe	ADJ
ap-948	183	9	,	,	PUNCT
ap-948	183	10	b.	b.	PROPN
ap-948	183	11	,	,	PUNCT
ap-948	183	12	schiele	schiele	PROPN
ap-948	183	13	,	,	PUNCT
ap-948	183	14	b.	b.	PROPN
ap-948	183	15	:	:	PUNCT
ap-948	183	16	local	local	ADJ
ap-948	183	17	features	feature	NOUN
ap-948	183	18	for	for	ADP
ap-948	183	19	object	object	NOUN
ap-948	183	20	class	class	NOUN
ap-948	183	21	recognition	recognition	NOUN
ap-948	183	22	.	.	PUNCT
ap-948	184	1	technical	technical	ADJ
ap-948	184	2	report	report	NOUN
ap-948	184	3	,	,	PUNCT
ap-948	184	4	multimodal	multimodal	NOUN
ap-948	184	5	interactive	interactive	ADJ
ap-948	184	6	systems	system	NOUN
ap-948	184	7	tu	tu	PROPN
ap-948	184	8	darmstadt	darmstadt	PROPN
ap-948	184	9	,	,	PUNCT
ap-948	184	10	germany	germany	PROPN
ap-948	184	11	,	,	PUNCT
ap-948	184	12	2005	2005	NUM
ap-948	184	13	.	.	PUNCT
ap-948	185	1	[	[	X
ap-948	185	2	10	10	NUM
ap-948	185	3	]	]	X
ap-948	185	4	mikolajczyk	mikolajczyk	NOUN
ap-948	185	5	,	,	PUNCT
ap-948	185	6	k.	k.	PROPN
ap-948	185	7	,	,	PUNCT
ap-948	185	8	tuytelaars	tuytelaar	NOUN
ap-948	185	9	,	,	PUNCT
ap-948	185	10	t.	t.	PROPN
ap-948	185	11	,	,	PUNCT
ap-948	185	12	schmid	schmid	PROPN
ap-948	185	13	,	,	PUNCT
ap-948	185	14	c.	c.	PROPN
ap-948	185	15	,	,	PUNCT
ap-948	185	16	zisserman	zisserman	NOUN
ap-948	185	17	,	,	PUNCT
ap-948	185	18	a.	a.	NOUN
ap-948	185	19	,	,	PUNCT
ap-948	185	20	matas	matas	PROPN
ap-948	185	21	,	,	PUNCT
ap-948	185	22	j.	j.	PROPN
ap-948	185	23	,	,	PUNCT
ap-948	185	24	schaffalitzky	schaffalitzky	PROPN
ap-948	185	25	,	,	PUNCT
ap-948	185	26	f.	f.	PROPN
ap-948	185	27	,	,	PUNCT
ap-948	185	28	van	van	PROPN
ap-948	185	29	gool	gool	PROPN
ap-948	185	30	,	,	PUNCT
ap-948	185	31	l.	l.	PROPN
ap-948	185	32	,	,	PUNCT
ap-948	185	33	kadir	kadir	PROPN
ap-948	185	34	,	,	PUNCT
ap-948	185	35	t.	t.	PROPN
ap-948	185	36	:	:	PUNCT
ap-948	185	37	a	a	DET
ap-948	185	38	comparison	comparison	NOUN
ap-948	185	39	of	of	ADP
ap-948	185	40	affine	affine	NOUN
ap-948	185	41	region	region	NOUN
ap-948	185	42	detectors	detector	NOUN
ap-948	185	43	.	.	PUNCT
ap-948	186	1	international	international	ADJ
ap-948	186	2	journal	journal	NOUN
ap-948	186	3	of	of	ADP
ap-948	186	4	computer	computer	NOUN
ap-948	186	5	vision	vision	NOUN
ap-948	186	6	,	,	PUNCT
ap-948	186	7	2005	2005	NUM
ap-948	186	8	.	.	PUNCT
ap-948	187	1	[	[	X
ap-948	187	2	11	11	NUM
ap-948	187	3	]	]	X
ap-948	187	4	ter	ter	NOUN
ap-948	187	5	haar	haar	PROPN
ap-948	187	6	romeny	romeny	PROPN
ap-948	187	7	,	,	PUNCT
ap-948	187	8	b.	b.	NOUN
ap-948	187	9	,	,	PUNCT
ap-948	187	10	florack	florack	NOUN
ap-948	187	11	,	,	PUNCT
ap-948	187	12	l.	l.	PROPN
ap-948	187	13	,	,	PUNCT
ap-948	187	14	koenderink	koenderink	PROPN
ap-948	187	15	,	,	PUNCT
ap-948	187	16	j.	j.	PROPN
ap-948	187	17	,	,	PUNCT
ap-948	187	18	viergever	viergever	ADV
ap-948	187	19	,	,	PUNCT
ap-948	187	20	m.	m.	NOUN
ap-948	187	21	(	(	PUNCT
ap-948	187	22	editors	editor	NOUN
ap-948	187	23	):	):	PUNCT
ap-948	187	24	scale	scale	NOUN
ap-948	187	25	-	-	PUNCT
ap-948	187	26	space	space	NOUN
ap-948	187	27	theory	theory	NOUN
ap-948	187	28	in	in	ADP
ap-948	187	29	computer	computer	NOUN
ap-948	187	30	vision	vision	NOUN
ap-948	187	31	,	,	PUNCT
ap-948	187	32	first	first	ADJ
ap-948	187	33	international	international	ADJ
ap-948	187	34	conference	conference	NOUN
ap-948	187	35	,	,	PUNCT
ap-948	187	36	scale	scale	NOUN
ap-948	187	37	-	-	PUNCT
ap-948	187	38	space’97	space’97	NOUN
ap-948	187	39	,	,	PUNCT
ap-948	187	40	utrecht	utrecht	PROPN
ap-948	187	41	,	,	PUNCT
ap-948	187	42	the	the	DET
ap-948	187	43	netherlands	netherlands	PROPN
ap-948	187	44	,	,	PUNCT
ap-948	187	45	july	july	PROPN
ap-948	187	46	2–4	2–4	NUM
ap-948	187	47	,	,	PUNCT
ap-948	187	48	1997	1997	NUM
ap-948	187	49	,	,	PUNCT
ap-948	187	50	proceedings	proceeding	NOUN
ap-948	187	51	.	.	PUNCT
ap-948	188	1	springer	springer	NOUN
ap-948	188	2	,	,	PUNCT
ap-948	188	3	1997	1997	NUM
ap-948	188	4	.	.	PUNCT
ap-948	189	1	[	[	X
ap-948	189	2	12	12	NUM
ap-948	189	3	]	]	PUNCT
ap-948	189	4	turk	turk	PROPN
ap-948	189	5	,	,	PUNCT
ap-948	189	6	m.	m.	NOUN
ap-948	189	7	,	,	PUNCT
ap-948	189	8	pentland	pentland	PROPN
ap-948	189	9	,	,	PUNCT
ap-948	189	10	a.	a.	NOUN
ap-948	189	11	:	:	PUNCT
ap-948	189	12	eigenfaces	eigenface	NOUN
ap-948	189	13	for	for	ADP
ap-948	189	14	recognition	recognition	NOUN
ap-948	189	15	.	.	PUNCT
ap-948	190	1	journal	journal	PROPN
ap-948	190	2	of	of	ADP
ap-948	190	3	cognitive	cognitive	ADJ
ap-948	190	4	neuroscience	neuroscience	NOUN
ap-948	190	5	,	,	PUNCT
ap-948	190	6	vol	vol	NOUN
ap-948	190	7	.	.	NOUN
ap-948	190	8	3	3	NUM
ap-948	190	9	(	(	PUNCT
ap-948	190	10	1991)no	1991)no	NUM
ap-948	190	11	.	.	PROPN
ap-948	190	12	1	1	NUM
ap-948	190	13	,	,	PUNCT
ap-948	190	14	p.	p.	NOUN
ap-948	190	15	71–86	71–86	NUM
ap-948	190	16	.	.	PUNCT
ap-948	191	1	[	[	X
ap-948	191	2	13	13	NUM
ap-948	191	3	]	]	X
ap-948	191	4	viola	viola	NOUN
ap-948	191	5	,	,	PUNCT
ap-948	191	6	p.	p.	PROPN
ap-948	191	7	,	,	PUNCT
ap-948	191	8	jones	jones	PROPN
ap-948	191	9	,	,	PUNCT
ap-948	191	10	m.	m.	NOUN
ap-948	191	11	:	:	PUNCT
ap-948	191	12	rapid	rapid	ADJ
ap-948	191	13	object	object	NOUN
ap-948	191	14	detection	detection	NOUN
ap-948	191	15	using	use	VERB
ap-948	191	16	a	a	DET
ap-948	191	17	boosted	boosted	ADJ
ap-948	191	18	cascade	cascade	NOUN
ap-948	191	19	of	of	ADP
ap-948	191	20	simple	simple	ADJ
ap-948	191	21	features	feature	NOUN
ap-948	191	22	.	.	PUNCT
ap-948	192	1	conference	conference	NOUN
ap-948	192	2	on	on	ADP
ap-948	192	3	computer	computer	NOUN
ap-948	192	4	vision	vision	NOUN
ap-948	192	5	and	and	CCONJ
ap-948	192	6	pattern	pattern	NOUN
ap-948	192	7	recognition	recognition	NOUN
ap-948	192	8	2001	2001	NUM
ap-948	192	9	,	,	PUNCT
ap-948	192	10	2001	2001	NUM
ap-948	192	11	.	.	PUNCT
ap-948	193	1	[	[	X
ap-948	193	2	14	14	NUM
ap-948	193	3	]	]	SYM
ap-948	193	4	yang	yang	PROPN
ap-948	193	5	,	,	PUNCT
ap-948	193	6	m.-h	m.-h	PROPN
ap-948	193	7	.	.	PROPN
ap-948	193	8	,	,	PUNCT
ap-948	193	9	kriegman	kriegman	PROPN
ap-948	193	10	,	,	PUNCT
ap-948	193	11	d.	d.	PROPN
ap-948	193	12	j.	j.	PROPN
ap-948	193	13	,	,	PUNCT
ap-948	193	14	ahuja	ahuja	PROPN
ap-948	193	15	,	,	PUNCT
ap-948	193	16	n.	n.	NOUN
ap-948	193	17	:	:	PUNCT
ap-948	193	18	detecting	detect	VERB
ap-948	193	19	faces	face	NOUN
ap-948	193	20	in	in	ADP
ap-948	193	21	images	image	NOUN
ap-948	193	22	:	:	PUNCT
ap-948	193	23	a	a	DET
ap-948	193	24	survey	survey	NOUN
ap-948	193	25	.	.	PUNCT
ap-948	194	1	pattern	pattern	NOUN
ap-948	194	2	analysis	analysis	NOUN
ap-948	194	3	and	and	CCONJ
ap-948	194	4	machine	machine	NOUN
ap-948	194	5	intelligence	intelligence	NOUN
ap-948	194	6	,	,	PUNCT
ap-948	194	7	ieee	ieee	NOUN
ap-948	194	8	transactions	transaction	NOUN
ap-948	194	9	on	on	ADP
ap-948	194	10	,	,	PUNCT
ap-948	194	11	vol	vol	NOUN
ap-948	194	12	.	.	PROPN
ap-948	194	13	24	24	NUM
ap-948	194	14	(	(	PUNCT
ap-948	194	15	2002	2002	NUM
ap-948	194	16	)	)	PUNCT
ap-948	194	17	,	,	PUNCT
ap-948	194	18	no	no	INTJ
ap-948	194	19	.	.	NOUN
ap-948	194	20	1	1	NUM
ap-948	194	21	,	,	PUNCT
ap-948	194	22	p.	p.	NOUN
ap-948	194	23	34–58	34–58	NUM
ap-948	194	24	.	.	PUNCT
ap-948	195	1	[	[	X
ap-948	195	2	15	15	NUM
ap-948	195	3	]	]	X
ap-948	195	4	zhang	zhang	PROPN
ap-948	195	5	,	,	PUNCT
ap-948	195	6	j.	j.	PROPN
ap-948	195	7	,	,	PUNCT
ap-948	195	8	yan	yan	PROPN
ap-948	195	9	,	,	PUNCT
ap-948	195	10	y.	y.	PROPN
ap-948	195	11	,	,	PUNCT
ap-948	195	12	lades	lade	NOUN
ap-948	195	13	,	,	PUNCT
ap-948	195	14	m.	m.	NOUN
ap-948	195	15	:	:	PUNCT
ap-948	195	16	face	face	NOUN
ap-948	195	17	recognition	recognition	NOUN
ap-948	195	18	:	:	PUNCT
ap-948	195	19	eigenface	eigenface	NOUN
ap-948	195	20	,	,	PUNCT
ap-948	195	21	elastic	elastic	ADJ
ap-948	195	22	matching	matching	NOUN
ap-948	195	23	and	and	CCONJ
ap-948	195	24	neural	neural	ADJ
ap-948	195	25	nets	net	NOUN
ap-948	195	26	.	.	PUNCT
ap-948	196	1	proceedings	proceeding	NOUN
ap-948	196	2	of	of	ADP
ap-948	196	3	the	the	DET
ap-948	196	4	ieee	ieee	NOUN
ap-948	196	5	,	,	PUNCT
ap-948	196	6	vol	vol	NOUN
ap-948	196	7	.	.	PROPN
ap-948	196	8	85	85	NUM
ap-948	196	9	(	(	PUNCT
ap-948	196	10	1997	1997	NUM
ap-948	196	11	)	)	PUNCT
ap-948	196	12	,	,	PUNCT
ap-948	196	13	no	no	INTJ
ap-948	196	14	.	.	NOUN
ap-948	196	15	9	9	X
ap-948	196	16	.	.	X
ap-948	197	1	peter	peter	PROPN
ap-948	197	2	hosten	hosten	PROPN
ap-948	197	3	e	e	NOUN
ap-948	197	4	-	-	NOUN
ap-948	197	5	mail	mail	NOUN
ap-948	197	6	:	:	PUNCT
ap-948	197	7	hosten@ient.rwth-aachen.de	hosten@ient.rwth-aachen.de	ADJ
ap-948	197	8	ing	ing	PROPN
ap-948	197	9	.	.	PUNCT
ap-948	198	1	mark	mark	PROPN
ap-948	198	2	asbach	asbach	PROPN
ap-948	198	3	e	e	NOUN
ap-948	198	4	-	-	NOUN
ap-948	198	5	mail	mail	NOUN
ap-948	198	6	:	:	PUNCT
ap-948	198	7	asbach@ient.rwth-aachen.de	asbach@ient.rwth-aachen.de	PROPN
ap-948	198	8	institute	institute	PROPN
ap-948	198	9	of	of	ADP
ap-948	198	10	communications	communication	NOUN
ap-948	198	11	engineering	engineering	NOUN
ap-948	198	12	rwth	rwth	PROPN
ap-948	198	13	aachen	aachen	PROPN
ap-948	198	14	university	university	PROPN
ap-948	198	15	52056	52056	NUM
ap-948	198	16	aachen	aachen	PROPN
ap-948	198	17	,	,	PUNCT
ap-948	198	18	germany	germany	PROPN
ap-948	198	19	©	©	PROPN
ap-948	198	20	czech	czech	PROPN
ap-948	198	21	technical	technical	PROPN
ap-948	198	22	university	university	PROPN
ap-948	198	23	publishing	publishing	NOUN
ap-948	198	24	house	house	NOUN
ap-948	198	25	http://ctn.cvut.cz/ap/	http://ctn.cvut.cz/ap/	PROPN
ap-948	198	26	7	7	NUM
ap-948	198	27	acta	acta	PROPN
ap-948	198	28	polytechnica	polytechnica	PROPN
ap-948	198	29	vol	vol	NOUN
ap-948	198	30	.	.	PUNCT
ap-948	199	1	47	47	NUM
ap-948	199	2	no	no	NOUN
ap-948	199	3	.	.	PUNCT
ap-948	200	1	4–5/2007	4–5/2007	NUM