Inernaional Associaion of Scienific Innovaion and Research (IASIR) (An Associaion Unifying he Sciences, Engineering, and Applied Research) Inernaional Journal of Emerging Technologies in Compuaional and Applied Sciences (IJETCAS) www.iasir.ne ISSN (Prin): 2279-0047 ISSN (Online): 2279-0055 Evaluaion of he Digial images of Penaeid Prawns Species Using Canny Edge Deecion and Osu Thresholding Segmenaion V.Sucharia 1, S.Jyohi 2,D.M.Mamaha 3 1 Dep of Compuer science & Engg, ASCET, Gudur, INDIA 2 Dep of compuer science, S.P.M.V.V, Tirupai, INDIA 3 Dep of Sericulure, S.P.M.V.V, Tirupai, INDIA Absrac: Image segmenaion is a key opic of research from many years. Image segmenaion plays a very imporan role in compuer vision.the purpose of image segmenaion is o pariion he image ino an se of disjoin regions wih he homogeneous and uniform aribues like inensiy, one, color and exure. Edge deecion in images considerably reduces he amoun of daa and filers useless informaion, while considering he imporan properies of an image. There are various mehods for he segmenaion of he image. In his paper he segmenaion algorihms are analyzed using Canny edge deecion and Osu hresholding and are esed wih differen species of Prawn images. The paper focuses mainly on he effeciveness and efficiency of he wo echniques, used for esing heir suiabiliy for he ype of Prawn images. Keywords: Image segmenaion, Image analysis, Canny edge deecion, Thresholding, Prawn species I. Inroducion Prawn species classificaion and recogniion is an acive research area in aquaculure. Many feaures may be differen among he differen species of he prawn. The classificaion is made by analyzing he feaures of he prawn. To exrac he feaures of he prawn Image segmenaion has o be done because segmenaion is he firs and foremos sep of image processing applicaions in which he properies of objecs in an image needs o be analyzed. The image segmenaion is an imporan ask for he exracion of useful informaion from he image such as he color, shape, exure and he srucure. Image segmenaion is he process of pariioning an image ino differen segmens. The main goal of segmenaion is o idenify some objecs of ineres depiced in he image so ha we can simplify or he change he represenaion of he image ino more meaningful form o analyze in easy manner. Based on he image acquisiion ype he Images can be classified ino differen ypes. The mos common ype of images which we see are ligh inensiy images, hey represen he variaion of ligh inensiy. The common digial image processing asks are Zooming, image segmenaion, resizing, edge deecion and color enhancemen. Among hese echniques, his paper focused on image segmenaion of digial images. Image segmenaion is he process of giving a label o each and every pixel of an image so ha he pixels wih same label share some feaures. In an image segmenaion he segmens ha cover he enire image or se of conours exraced from he image. In a paricular region some pixels are same based on color, exure or inensiy [1],[2] Image segmenaion is usually done using edge deecion echniques which are basically 2-D filers and deecs he edges depending upon he level of inensiy difference in pixels and he level of disconinuiy. The effecive image segmenaion is very difficul and challenging ask in he processing of an image [11],[12]. The Segmenaion algorihms ha are based on disconinuiy approach pariions an image based on rapid changes in inensiy and hose on similariy approach are based on pariioning an image ino regions ha are similar based on he predefined crieria [3],[4],[5]. As a resul of image segmenaion more meaningful image is obained which is easier o undersand. Because images provide semanically poor informaion, image segmenaion is essenially an applicaion oriened problem ha demands he involvemen of he human expers or applicaion specific soluions. The choice of he paricular segmenaion echnique depends on he ype of ask o be performed and he naure of he images available. II. Segmenaion Algorihms The basic sep is o segmen an image in image analysis. There are so many segmenaion algorihms. For he inensiy images he popular segmenaion mehods are Edge deecion mehods, hresholding based, Region based and conneciviy preserving relaxaion mehods. The edge based mehods hey cener around he conour deecion. Threshold mehods ake decisions based on he local informaion of pixels and hey are effecive if he inensiy levels of he objecs fall ouside he range of he levels in he background. In he region based mehods he image is pariioned ino he regions conneced by grouping neighboring pixels of same inensiy levels. Segmenaion based on he disconinuiy of he edges or segmenaion based on Region which segmens an image ino regions based on he likeness according o a crieria predefined, in his paper mainly he wo mehods are discussed and evaluaed, hey are edge based deecion echnique and hreshold based segmenaion. Canny edge deecion is seleced for edge based segmenaion and Osu hresholding segmenaion represen he oher mehod. IJETCAS 13-526; 2013, IJETCAS All Righs Reserved Page 117
V. Sucharia e al., Inernaional Journal of Emerging Technologies in Compuaional and Applied Sciences, 6(2), Sepember-November, A. Segmenaion by Edge Deecion Technique The Segmenaion Mehods based which are based on disconinuiy search for quick changes in he inensiy value are called edge based mehods[1],[7]. Edge deecion echniques are used for finding disconinuiies in inensiy values. The edge is he boundary in beween wo regions wih differen grey level feaures. Edge based segmenaion mehods deec he disconinuiies and produce he binary images conaining edges and also heir background as he oupu of hem. Imporan feaures can be exraced from he edges.[3] Edge deecion is used for objec recogniion and so many oher applicaions. Edges can be deeced in differen ways such as Sobel, Robers, Prewi and Canny operaors. The Canny edge deecion is presened here. Canny Edge Deecor The Canny edge deecor [12] is deermined as he bes edge deecors, Canny's edge deecor ensures good noise immuniy and a he same ime deecs rue edge poins wih minimum error. Canny mehod has opimized he edge deecion wih respec o cerain crieria. The firs wo of hese crieria discusses he issue of deecion. If an edge is given wheher he edge deecor will find he edge or no. The hird crieria address issue of he localizaion meaning ha how exacly he posiion of an edge is idenified. The seps of he Canny algorihm are as follows: 1. Smoohing of an image: Blurring of he image o eliminae noise by using he Gaussian filer. 2. Finding he gradiens: The edges should be marked where he gradiens of image has large magniudes. 3. Non-maximum suppression: Only local maxims should be marked as edges. finds he local maxima in he direcion of he gradien, and suppresses all ohers, minimizing false edges. 4. Double hresholding: Poenial edges are deermined by hresholding, Insead of using a single hreshold value for he enire image, he hyseresis hresholding is used by Canny algorihm which has some adapiviy o he local conen of he image. There are wo hreshold levels, h, high and l, low where h > l. Pixel values above he h value wihou delay are classified as edges. 5. Edge racking by hyseresis: Final he edges are deermined by suppressing all edges which are no conneced o a definie edge. B. Thresholding Segmenaion Technique The hresholding mehods are generally used o segmen an images ino various classes, which consis of dark objec and brigh background or a brigh objec and dark background[4],[6],[10]. The image segmenaion depends upon he hreshold value which changes according o he feaure values of an image. The gray scale or color images are segmened based on depending he gray values which conver he color or gray scale images ino he binary images by aking ino accoun each pixel[8]. The inpu o he hresholding mehod is a color image or gray scale image. The oupu is he binary image represening he image segmenaion. Thresholding creaes he binary images from grey level ones by ransforming all pixels below some hreshold o zero and all he pixels above o one. Thresholding is he ransforming an inpu image o oupu image g as shown in he formula. Where g(x,y) is hresholded version of f(x,y) a hreshold value. where is he hreshold, g(i, j) = 1 for image elemens as he objecs, and g(i, j) = 0 for image elemens as he background, in his way he chosen hreshold is dependable on he grey level value, his is called global hresholding echnique. The hresholding echnique is mainly divided ino hree caegories: (i) Local Thresholding In he local hresholding mehod he parameers of hreshold is considered over a small area[9]. (ii) Adapive Thresholding or Dynamic Thresholding If T is depending on he spaial co-ordinaes hen he hresholding is called as adapive hresholding or dynamic hresholding. In his ype differen hreshold is used for differen regions in he image. The hreshold changes dynamically over image. If he value of he pixel is below he hreshold value hen i is se o he background value else i se as a foreground value. (iii) Global Thresholding If he inensiy disribuion and background pixel are differen hen gobal hresholding mus be used on he enire image. If T depends on he grey level value of he image and T is exclusively relaed o he properies of pixel in image, hen his mehod is called global hresholding. Osu mehod is called as a Global hresholding mehod. Osu Thresholding A measure of region homogeneiy is variance ha is he regions wih high homogeneiy will have he low variance. Osu s hresholding echnique selecs he hreshold by minimizing wihin-class variance of wo groups of pixels separaed by he hresholding operaor[12]. Osu hreshold echnique is used in many applicaions from image analysis o compuer vision. I does no depend on modeling he probabiliy densiy funcions, on he oher hand, i assumes a bimodal disribuion of gray-level values ha if he image approximaely fis he condiion i will do a nice job. Osu is based on he hreshold for pariioning he pixels IJETCAS 13-526; 2013, IJETCAS All Righs Reserved Page 118
V. Sucharia e al., Inernaional Journal of Emerging Technologies in Compuaional and Applied Sciences, 6(2), Sepember-November, of an image ino wo classes C 0 and C 1 a grey level, where : C 0 = {1, 1, 2,, } and C 1 = { + 1, +2,.l - 1}, and le q 0 and q 1 and represen he esimae of class probabiliies defined as follows: q0( ) p( i), and l 1 q1 ( ) p( i) i 0 1 and sigmas are he individual class variances defined by: pi () ( ) [ i ( )], and q () i 2 2 0 0 i 0 0 l 1 2 2 pi () 1 ( ) [ i 1( )] 1 q1 () Where he class means are defined by: l 1 ip() i ip() i 0() 1() i 0 q0() 1 q, and 1() Here, P represens he hisogram of he image. The problem of minimizing wihin he class variance can be expressed as maximizing beween class variance which can be shown as a difference of oal variance and wihin class variance: 2 2 2 ( ) ( ) q ( )[1 q ] b w 0 0( ) [ ( ) ( )] 1 0 2 This expression can be maximized and he soluion is he ha maximizing The seps of he algorihm are as follows: 1.Calculae he hisogram and he probabiliies of he each inensiy level. 2.Se up iniial q i (0) and μ i (0). 3.Sep hrough all possible hresholds o maximum inensiy. Updae q i and μ i. 4.Calculae () 5.The preferred hreshold corresponds o he maximum. () (). III. Experimenal Verificaions The paper mainly presens wo echniques of image segmenaion one is Canny edge deecion and he oher is Osu hresholding, hey are esed wih a various species of prawn images and heir corresponding segmenaion using he wo mehods, as examples of our experimens. The original images of hree species of he prawn namely Penaeus Vannamei, Penaeus Monodon and Penaeus Indicus as shown in Figure 1. A. Tesing Procedure The edge deecion segmenaion was implemened using MATLAB and esed for he species of prawn images. Figure 1. Prawn images B. Simulaion Resuls The canny edge deecion and osu hresholding is implemened. An user inerface was designed for he selecion of he image o perform canny and osu as shown in Figure 2.The performance resuls applied by he wo Techniques canny and osu hresholding are shown in he Figure 3 and Figure 4. IJETCAS 13-526; 2013, IJETCAS All Righs Reserved Page 119
V. Sucharia e al., Inernaional Journal of Emerging Technologies in Compuaional and Applied Sciences, 6(2), Sepember-November, Figure 2 : Screen sho Figure 3. Canny Edge Deecion Figure 4. Osu segmened images IV. Conclusion In image based analysis he segmenaion is pariioning of he digial image ino muliple regions which is a se of pixels, based on some crieria. The problem of he segmenaion wih various approaches is sudied clearly for he idenificaion of he prawn species. Differen ypes of approaches are suied o various ypes of images and he qualiy oupu of a paricular algorihm is no easy o measure because here may be so many correc segmenaions for a single image. In his paper he effeciveness of he algorihms are evaluaed for hree differen species of penaeid prawn images like Vannamei, Monodon and Indicus as shown in he figure.the Segmened Images are good ha are obained by boh he ypes of he algorihms. The resuls given by he canny are quie good. The resul gives us an idea abou he efficiency of he algorihms are responsible based on he ype of images and heir applicaions. In his paper for he prawn images he effeciveness of he algorihms ha are proposed are evaluaed as seen in Fig.1. Boh he algorihms give nice segmened images, bu for he objecs ha are noable from background in he images Osu is more suiable. Canny segmenaion is more suiable han Osu o he Prawn images because all he edges are clearly idenified from he background so ha we can exrac he edge feaures as shown in Figure 2. I is recommended for he fuure work o une he Canny parameers like hreshold and sigma o give more effecive resuls and o deal wih a variey of images. V. References [1] R. C. Gonzalez and R. E. Woods. Digial Image Processing, Second Ediion, Prenice Hall, 2002. [2] Y.Zhang, A survey on evaluaion mehods for image segmenaion [3] B. Chanda., D. D. Majumder. Digial Image Processing and Analysis, Prenice Hall, 2003. [4] L. Spirkovsk. A Summary of Image Segmenaion Techniques, Ames Research Cener, Moffe Field, California, 1993. IJETCAS 13-526; 2013, IJETCAS All Righs Reserved Page 120
V. Sucharia e al., Inernaional Journal of Emerging Technologies in Compuaional and Applied Sciences, 6(2), Sepember-November, [5] J. Saif and A. Moharram, Edge and Region Based Image Segmenaion, Journal of Compuer and Informaion Technology of Hodeidah Universiy, Hodedah, Yemen, 2011. [6] E. A. Savakis. Adapive Documen Image Thresholding Using Foreground and Background clusering, published in Proceeding of Inernaional Conference on Image Processing ICIP, 98 [7] P. Thakare. A Sudy of Image Segmenaion and Edge Deecion Techniques, Inernaional Journal on Compuer Science and Engineering(IJCSE), Feb. 2011. [8] Liu Ping, A Survey on hreshold selecion of image segmenaion, journal of image and graphics,2004,pp86-92. [9] Zhang, Y,J.,1997. Evaluaion and comparision of differen segmenaion algorihms. Paern recogniion leers 18(!0).pp.963-974 [10] Haralick, R.M. and Shapiro, L.G. 1985.Image segmenaion echniques. Compuer vision,graphics and Image processing,29(1)pp.100-132. [11] J.F Canny, A Compuaional approach o edge deecion, IEEE rans. Paern Analysis and Machine Inelligence,8:679-698,1986. [12] Jamil.A.M. Saif, Ali Abdo Mohammad Al-Kubai, Abdul awab saif hazaa, Mohammad al moraish, Image Segmenaion using Edge Deecion and Thresholding, ACIT2012 Dec 10-13 ISSN:1812-0857 VI. Acknowledgemens Auhors would like o hank DBT, New Delhi for sancioning he Projec. Currenly his work is carried ou under DBT Projec. IJETCAS 13-526; 2013, IJETCAS All Righs Reserved Page 121