ة مجلة إبن الھیثم للعلوم الصرفة و التطبیقی 2012 السنة 25 المجلد 1 العدد Ibn A l-Haitham Journal f or Pure and Applied Science No. 1 Vol. 25 Year 2012 The Effect of Window Size Changing on Satellite Image Segmentation Using 2D Fast Otsu Method T. A. Naji Departme nt of Physics, College of Education I bn-Al-Haithem, Unive rsity of Baghdad Received in: 17June 2011,Accepted in: 20Septe mber 2011 Abstract M ultisp ectral remote sensin g image se gmentation can be achieved usin g a multithresholding technique. This p aper studies t he effect of chan ging the window size of the two dimensional (2D) fast Ot su algorithm that p resented by Zhang. From the results, it shown that this method behaves as a se arch machine for the valleys (an automatic threshold), between the gray levels of t he hist ogram with changing the size of slide window. Keywords I mage Segmentation, (2D) Fast Otsu method, M ultithresholding, Aut omatic thresholding, (2D) hist ogram image. Introduction Image se gmentation p lay s an imp ortant role in image analy sis and comp uter vision system, this is the process for subdividing an image into the homo geneous re gions. Among all segmentation techniques, the automatic t hresholdin g methods are widely used tools for image segmentation as a p reprocessing st ep, because of their advantages of si mple imp lement and time saving. Ot su's method is one of thresholding methods and frequently used in various field. This is used to automatically p erform histogram shape-based image thresholding, or the reduction of a gray level ima ge to a binary image. Two dimensional (2D) Ot su method behaves well in se gmentation images of low signal to noise ratio than one dimensional (1D), but it gives satisfactory results only when the numbers of p ixel in each class are close to each other [1, 2, 3, and 4]. This p aper st udies the effect of changing the slide window size of the (2D) fast Ot su algorithm, for ima ge se gmentation based on the value of central gray level of the highest frequency on the (2D) histogram image of the slide window selection. The changing of slide window size is set manually by the user and it dep ends on the content of the images. Samples of Test Scenes Three available sc enes for differ ent Iraq’s regions (shown in fig.(3 a, 5a, 6a)) whi ch comp rise 581.7 km 2 , have been chosen and used for testing the introduced segmentation algorithm; these are:  Al Rama di: located in Al-Anbar p rovenience west of Baghdad. The available scene was TM exp osure at M arch 04, 1990. This region lies between latitudes. 33° 27' 21.44" N to 33° 14' 14.05" N, and longitudes 43° 29' 17.91" E to 43° 44' 36.64" E. This region represents Alluvial Plain ; the hot desert c limate p revails in the sedimentary p lain and the west ern p lateau. It contains some vegetation cover, many soil and rock erosion noticed. ة مجلة إبن الھیثم للعلوم الصرفة و التطبیقی 2012 السنة 25 المجلد 1 العدد Ibn A l-Haitham Journal f or Pure and Applied Science No. 1 Vol. 25 Year 2012  Al Raz z az ah : located between Karbala and Ramadi p roveniences. This region lies between latitudes. 32° 47' 59.07" N t o 32° 34' 49.06" N, and longitudes 43° 47' 58.44" E to 44° 3' 4.22" E. The scene (TM taken at 1987) rep resents Alluvial Plain.  Al Fit’h a: situated north Salah al- Den province. This region lies between latitudes 35° 20' 26.29" N to 35° 6' 45.46", and longitudes 43° 18' 38.39" E to 43° 35' 9.45" E. The general features are the merge of the lower-Zab with the Tigris Rivers, and vegetation cover can be detected. This image was ETM + exp osure at M arch 19, 2002. The map location of these scenes was shown in fig. (1). Algorithm To study the effect of window size changing usin g (2D) fast Otsu method, the following st ep s can be p erformed: 1) App ly ing the (2D) fast Otsu method on the test images using d ifferent sizes for the slide windows. 2) In each slide window, two variables are comp uted; the central gray lev el (GL) of the slid e window and the slid e window local mean for the high est frequency in the (2D) histogram image. 3) Plot between the computed central gray level from st ep (2), and window size. 4) Find the flat area from the curve, because it represents the op timum threshold value (valley ). Experime ntal Re sults In order to study the influence of incr easin g size of slide window of the (2D) fast Otsu algorithm that p resented by Zhang. Three different regions of Iraq are used which has the (827×866) p ixel with (256) gray level intensities, the choice of these study area was influenced by the variety of sp atial and ecological land. This st udy was p erformed using 64- bit computer p latform of core 2 Due 2.2GHz p rocesser and M ATLAB Ver. 2009a lan guage. The new histogram had counted usin g sl ide window with two variables which are the central gray level (GL), and local mean of the slide window, as a (2D) h istogram image. This (2D) hist ogram used to the extension of the Ot su t hreshold algorithm p ossibility . The highest frequency on (2D) histogram ima ge of the sl ide window which represent the p ossible sp lit p oint (threshold value) of the central gray level. For examp le, fig. (2) illust rates the (2D) histogram v ia (3D) fe ature sp ace plot with window size (50×50) p ixel, the maximum p eak at central gray level and local mean are 120, 114 resp ectively, where the (x-axis) and (y-axis) represent t he central gray level (GL) of the slide window, and the slide window local mean resp ectively. The increasin g of window size is st arted from less value as 5 to higher v alue as 400. When the small window size has been chosen (as 5, 15, and 25), the small difference between the central gr ay level (GL) and lo cal mean of the slide window have been obtained. ة مجلة إبن الھیثم للعلوم الصرفة و التطبیقی 2012 السنة 25 المجلد 1 العدد Ibn A l-Haitham Journal f or Pure and Applied Science No. 1 Vol. 25 Year 2012 This is due to the small window size mean small land coverage which mean less land features and mor e homo geneously in the slide window, and viceversa. The central gray lev el (GL) and lo cal mean values ar e similar (as Al Razz azah image, take window size 5), and are near (as Al-Ramadi and Al Fit’ha images, take window size 5, 15, and 25) as shown in In order to determine the op timum threshold value (valley ) a p lot is drawn between the p ossible threshold values t hat is found using the (2D) histogr am with the window size, where the flat area rep resent t he optimum threshold value (valley ) to sep arate the regions (classes). This valley was selected from the windows t hat have stability value of the central gray lev el (GL), the under lined stability value is shown in table (1). The curves are v erified in f ig. (3c), fig. (5c), and f ig. (6c) showed a true number of valleys between the gray levels of the histogram for extracting the objects from their background. In order to compare b etween the fast Ot su and the (2D) fast Otsu methods, the threshold values selection (valleys) for the tested scenes using fast Otsu method are list ed in table (2). Two valley s from Al Ramadi scene has been deriv ed, so that fig. (4) illustrates the classification of this scene with t hese valleys implementing. Conclusions From the obtained result, t he following conclusions can be derived: 1- The (2D) fast Otsu algorithm can b e used as search machine for the valley between the histogram p eaks, as shown in fig. (3b, 3 c), fig. (5b, 5c), and fig. (6b, 6c). 2- Changin g the window size in the (2D) fast Otsu algorithm has contrarily relationship with the gray level of the histogr am. Starting with small window size will for ce the algorithm to search the high gray level ar ea in the histogram, while takin g larger window size will slide the searching area for a valley in to smaller gr ay level in the hist ogram, as shown in fig. (3c), fig. (5c), and f ig. (6c). 3- If the algorithm finds a gray level for a v alley, changing the size of the slide window wi ll not change the found gray level until it finds another valley, as shown in fig. (3b, 3c). 4- It is seen from f ig. (3c) and fig. (6c), the d etected threshold gr ay level (valley) will appear as a flat curve. 5- From the comp arison between the result of the fast Otsu algorithm and the (2D) fast Ot su, it is found that there is no match between the threshold gray level values, as shown in table (2). 6- Scann ing the histogram with (2D) fast Otsu algorithm (by changing the window size) will give us the true number of valley in the histogram bett er than the other Otsu algorithms t hat force the algor ithm to sp lit the histogram to (n) number of bans even if 7- the histogram doesn't have enough number of valley to give an accurate sp litting bans (classification). ة مجلة إبن الھیثم للعلوم الصرفة و التطبیقی 2012 السنة 25 المجلد 1 العدد Ibn A l-Haitham Journal f or Pure and Applied Science No. 1 Vol. 25 Year 2012 Re ferences 1- Zhang, J. and Hu, J. (2008), Image se gmentation based on 2D Ot su method with histogram analysis, IEEE, International confer ence on computer sciences and soft ware engineering: 105. 2- Sahoo, P.K.; So ltani, S. and Wong, A.K.C. and Chen, Y.C. (1988), A survey of thresholding techniques, Computer Vision, Graphics, and Image processing, 41 : 233-260. 3- Ot su, N. (1979), A Threshold Selection M ethod from Gray -Level Hist ograms, IEEE Transactions on Sy stem, M an, and Cy bernetics, SM C-9 (1), January. 4- Sezgin, M . and Sankur, B. (2003), Survey over image thresholdin g techniques and quantitative performance evalu ation. Journal of Electronic Imaging, 13 (1): 146–165. Tabl e (1): Gray Le vel & Local Me an Values wi th Win dow Size Vari ati on Using 2D Fast O tsu Meth od Al-Ramadi Image Al-Razzazah I mage Al Fit’ha Image Window Size Gray Level Local Mean Gray Level Local Mean Gray Level Local Mean 5 121 118 113 113 91 94 15 120 116 102 104 81 83 25 120 115 97 100 80 85 50 120 114 89 97 78 85 55 120 114 86 96 78 84 75 120 113 80 89 75 91 100 119 112 77 88 75 97 125 72 78 70 97 75 105 150 72 82 66 101 67 107 187 72 84 60 107 54 94 210 72 87 60 107 48 100 250 72 91 54 108 39 112 300 69 94 48 108 28 123 350 69 98 44 108 17 129 400 66 103 43 107 9 134 ة مجلة إبن الھیثم للعلوم الصرفة و التطبیقی 2012 السنة 25 المجلد 1 العدد Ibn A l-Haitham Journal f or Pure and Applied Science No. 1 Vol. 25 Year 2012 Table (2): Th reshol d Values Using Fast Otsu Me th od Thresho lds Al-Ra madi Imag e Al-Ra zzaza h Image Al Fi t’ha Imag e 1 2 1 1 Fas t Otsu 72 15 4 13 2 12 9 Fi g. (1): Location Map of the Te st S cenes in ALRama di, Al Razz azah, an d Al Fi t’ Regions Al-Ramadi Al-Razzazah Al Fit’h Fi g. (2): 3D Fe atu re S pace Plot of 2D Fast Otsu Meth od with Win dow Size 50, the Maximum Peak at Central Gray Le ve l & Local Me an are 120, 114 Res pe cti vel y. ة مجلة إبن الھیثم للعلوم الصرفة و التطبیقی 2012 السنة 25 المجلد 1 العدد Ibn A l-Haitham Journal f or Pure and Applied Science No. 1 Vol. 25 Year 2012 a Fig. (3) :Al Ramadi Image, and its Histogram (a & b), (c) Window S ize Changing Curve with the Gray Level of Al Ramadi Image by Usi ng 2D Fast c b ة مجلة إبن الھیثم للعلوم الصرفة و التطبیقی 2012 السنة 25 المجلد 1 العدد Ibn A l-Haitham Journal f or Pure and Applied Science No. 1 Vol. 25 Year 2012 Fi g. (4): Cl assifie d Im age of Al Ramadi wi th 2 ة مجلة إبن الھیثم للعلوم الصرفة و التطبیقی 2012 السنة 25 المجلد 1 العدد Ibn A l-Haitham Journal f or Pure and Applied Science No. 1 Vol. 25 Year 2012 a Valley 1 b c Fig. (5): Al Razzazah Image, and its Histogram (a & b), (c) Window S ize Changing Curve with the Gray Level of Al Razzazah Image by Usi ng 2D Fast O tsu Method Valley 1 ة مجلة إبن الھیثم للعلوم الصرفة و التطبیقی 2012 السنة 25 المجلد 1 العدد Ibn A l-Haitham Journal f or Pure and Applied Science No. 1 Vol. 25 Year 2012 a c b Fig. (6): Al Fit’ha Image, and its Histogram (a & b), (c) Window S ize Changing Curve with the Gray Level of Al Fit’ha Image by Usi ng 2D Fast O tsu Method Valley 1 Valley 1 ة مجلة إبن الھیثم للعلوم الصرفة و التطبیقی 2012 السنة 25 المجلد 1 العدد Ibn A l-Haitham Journal f or Pure and Applied Science No. 1 Vol. 25 Year 2012 یر حجم النافذة لتقسیم الصورة الفضائیة بأستعمال طریقة أوتسو تأثیر تغ ِالسریعة الثنائیة األبعاد تغرید عبد الحمید ناجي قسم الفیزیاء، كلیة التربیة، جامعة بغداد 2011 ایلول 20: قبل البحث في ،2011 حزیران 17:استلم البحث في الخالصة درس هذا البحث تأثیر تغیر . عن بعد المتعددة االطیاف بأستعمال تقنیة حد العتبات تقسیم صور االستشعار ̋ یتم عادة كما هو معروض بالنتائج، تتصرف . زنك الباحث البعدین التي قدمت من قبليحجم النافذة لخوارزمیة أوتسو السریعة ذ للمخطط التكراري بتغیر حجم النافذة بین المستویات الرمادیة ) حد العتبة اآللي(هذه الطریقة كماكنة بحث عن الودیان .المنزلقة ط الكلمات المفتاحیة ِ تقسیم صورة، طریقة أوتسو السریعة الثنائیة األبعاد، حد العتبات، حد العتبة اآللي، صورة المخط .ِالتكراري الثنائي األبعاد ة مجلة إبن الھیثم للعلوم الصرفة و التطبیقی 2012 السنة 25 المجلد 1 العدد Ibn A l-Haitham Journal f or Pure and Applied Science No. 1 Vol. 25 Year 2012 ة مجلة إبن الھیثم للعلوم الصرفة و التطبیقی 2012 السنة 25 المجلد 1 العدد Ibn A l-Haitham Journal f or Pure and Applied Science No. 1 Vol. 25 Year 2012