# Introduction n image processing and retrievals image analysis plays a dominant role. The major task in image analysis is extracting useful information using features of the image. Generally image analysis techniques broadly grouping into groups namely (1) Structural methods (2) Statistical methods Raj Kumar et al (2011), among these two groups statistical methods are much popular. In Statistical methods one of the prime considerations is dividing whole image into different image regions using probability distributions. This type of method is usually referred as image Pal N. R. (1993), Cheng et al (2001), Srinivasa et al (2007), Srinivas Y et al (2010), Prasad Reddy et al (2007) have reviewed the image segmentation methods. There is no unique image segmentation method available for analyzing all images. The image segmentation is basically dependent on type of images. The image broadly categorized into four types of categories . They are (1) Images on Earth (2) images of Humans and animals (3) images on sky (4) images on Water and (5) images of Nature. Among these categories the images of Human beings and Animals are in different in nature and features are associated with these images are different from others in some statistical sense. These images are Skewed in nature. Hence the image segmentation methods based on Gaussian mixture model given by Cheng et al (2001), Yamazaki T. et al (1998), Zhang Z.H et al (2003), Lie T. et al (1993) may not suit well. Even the methods given by Sesha sayee et al (2011), Srinivasa et al (2011) are also may not suit since these methods also focus on symmetry of the pixel intensities in the image region. Hence to have suitable and more appropriate image segmentation methods for animals, an image segmentation method using a mixture of Pearsonian Type VI Distribution is developed and analyzed. Here it is assumed that whole image is characterized by a mixture of Pearsonian Type VI probability model. The Pearsonian Type VI Distribution is skewed in nature having long upper tails. This distribution also includes several distributions as particular case. From the Berkeley image data set collected over animal images. It is evident that the pixel intensities of these images are well categorized by mixture of Pearsonian Type VI Distribution. The model parameters are estimated by updated equations of EM algorithm. The initial values of the model parameters of EM Algorithm are carried using Histograms of the whole image and K-means and Hierarchical clustering Algorithm and moment method of estimates. The image segmentation algorithm is developed through Maximum Likelihood component under Bayesian frame. The performance of image segmentation algorithm is skewed using image quality metrics and ground truth values. The comparative study of proposed algorithm with that of Gaussian Mixture Model is also carried. Usually the entire image is considered as a union of several image regions in low level image analysis and the image data is quantified by pixel intensities in each image region. Because of the fact that the brightness measured at a point in the image is influenced by various random factors like environmental conditions, vision, moisture, lighting etc, the pixel intensity ( , ) z f x y = for a given point ( pixel ) (x, y) is a random variable. It is generally assumed that the pixel intensities of the region follow a Pearson Type VI distribution in order to model the pixel intensities of the animal and human image regions. The probability density function of the pixel intensity is ( ) ( ) 2 1 2 1 1 1 2 ( 1)1 2 2 ( / , , ) ) i i i i q q s i s i i i i q q i i i i i i z a z f z a q q a B q q q a z ? ? ? + ? = ? ? + ? < The entire animal and human image is a collection of regions which are characterized by Pearson Type VI distribution. Here, it is assumed that the pixel intensities of the whole image follows a Kcomponent mixture of Pearson Type VI distribution and its probability density function is of the form 1 1 2 ( ) ( / , , )(2) 1K i i i i i p z f z a q q i ? = ? = where, K is number of regions , 0 ? i ? ? 1 are weights such that ? i ? = 1 and 1 1 2 ( / , , ) i i i i f z a q q is as given in equation ( 1). In the whole image ? i is the weight associated with i th region. Usually the intensities of the pixel in the image regions are statistically correlated and can be reduced by spatial averaging ( Kelly P.A. et al ( 1998 ) ) or spatial sampling ( Lei T. and Sewehand W. ( 1992 )) .The pixels are considered to be uncorrelated and independent after reduction of correlation. The mean pixel intensity of the whole image is ( ) 1 i i K E Z i ? µ = ? = # III. Estimation of the Model Parameters by EM Algorithm In this section we derive the updated equations of the model parameters using Expectation Maximization (EM) algorithm. The likelihood function of the observations 1 2 , ,..., N z z z drawn from an image is N ( ) 1 ( ) ( , ) l s L p z s ? ? = = ? ( ) N That is ( ) ( , ) 1 1 ? ? = = = K L f z s i i i s ? ? ? ( ) Thisimplies log ( ) log ( , ) 1 1 ? ? = = = N K L f z s i i s i ? ? ? Where 1 1 2 ( , , , ; 1, 2,..., ) i i i i a q q i K ? ? = = is the set of parameters. # ( ) ( ) 2 1 2 1 ( 1) 1 2 2 log ( ) log ,(3)( ) ( 1, 1) 1 1 i i i i q q i s i s q q i i i i N K z a z L a B q q q s i ? ? ? ? + ? ? ? = ? ? ? ? ? ? + = = ? ? ? ? The first step of the EM algorithm requires the estimation of the likelihood function of the sample observations. # a) E-Step In the expectation (E) step, the expectation value of log ( ) L ? with respect to the initial parameter vector ? (0) is ( ) [ ] (0) ; log ( ) / (0) = Q z , belongs to any region K is ( ) ( ) ( ) ( ) ( , ) ( , ) ( ) ( , ) ( ) ( ) ( ) ( , )( , ) 1 == ? = ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? l l l l f z f z l s s k k k k t z s k K l l l p z f z Following the heuristic arguments of Jeff A. Bilmes (1997) we have ( ) ( ) ( ) ( ) ( ) ( ) ( )1 1 ; ( , ) log ( , ) log (5) = = = + ?? K N l l l l i s i s i i s Q t z f z ? ? ? ? ? But we have ( ) ( ) 2 1 2 1 1 1 2 ( 1) 1 2 2 ( / , , )( ) ( 1, 1)i i i i q q s i s i i i i q q i i i i z a z f z a q q a B q q q ? ? + ? = ? ? + ( D D D D ) Year 2014 F II. Mixture of Pearson Type vi Distribution ( , ) log ( , ) log 1 1 K N l l l l Q t z f z s s i i i i s ? ? ? ? ? ? ? = + = = b) M-Step For obtaining the estimation of the model parameters one has to maximize ( ) ( ) ; l Q ? ? such that ?? i = 1. This can be solved by applying the standard solution method for constrained maximum by constructing the first order Lagrange type function, ( ) ( ) ( ) (log ( )) 1 (6) 1 K l l S E L i i ? ? ? ? = + ? = ? ? ? ? ? ? where, ? is Lagrangian multiplier combining the constraint with the log likelihood function to be maximized. Hence, 0 ? = ? S i ? . This implies ( ) ( ) 2 1 2 1 ( ) i ( 1) 1 1 1 2 2 1 ( , ) log log ( ) ( 1, 1) 1 0 i i i i q q K N s i s l i s q q i s i i i i i K i i z a z t z a B q q q ? ? ? ? ? ? ? + = = = ? ? ? ? ? ? ? ? ? ? ? ? + ? ? ? ? ? + ? ? ? ? ? ? ? ? ? ? ? ? ? ? + ? = ? ? ? ? ?? ? This implies 1 1 ( ) ( , ) 0 i s i N l t z i ? ? ? + = ? = Summing both sides over all observations, we get ? = -N Therefore ( , ) ? = = ? N l i i s s t z N ? ? The updated equation of i ? for ( l +1) th iteration is ( ) 1 ( ) 1 1 ( , ) N l l i i s s t z N ? ? + = = ? ( ) 1 ( ) ( ) ( , )1( ) ( ) 1 ( , ) 1 l i l l N f z i i s K l l s N f z i i s i ? ? ? ? ? + ? = ? = ? ? ? ? = ? ? ? ?(7) Therefore ( ) ( ) ; 0 l i Q a ? ? ? = ? implies ( ) 1 log ( ; ) 0 l i L E a ? ? ? ? ? = ? ? ? ? ? ( ) ( ) ( ) ( ) ( )1 1 ( , ) log ( , ) log 0 K N l l l i s i s i i s i t z f z a ? ? ? = = ? ? ? ? ? + = ? ? ? ? ? # ?? The updated equation of i a at ( l +1)th iteration is ( ) 2 1 2 ( ) ( ) ( ) ( ) ( 1) ( ) ( ) 1 ( 1) ( , )( , )i i i i i l l l l N s i s l l l s i s q q z a t z a t z q ? ? + = ? ? ? ? + ? ? ? = ? ? ? ? ? For updating the parameter 1 i q , i = 1, 2, ?, K we consider the derivative of ( ) ( ) ; l Q ? ? with respect to 1 i q and equate it to zero. We have ( ) ( )( ) ; log ( ; ) l l Q E L ? ? ? ? ? ? = ? ? Therefore ( ) ( ) 1 ; 0 l i Q q ? ? ? = ? implies ( ) 1 log ( ; ) 0 l i L E q ? ? ? ? ? = ? ? ? ? ? ( ) ( ) ( ) ( ) ( ) 1 1 1 ( , ) log ( , ) log 0 K N l l l i s i s i i s i t z f z q ? ? ? = = ? ? ? + = ? ? ? ? ? ?? ( ) 1 1 ( ) 0 1 2 0 1 ( , ) 1 (8) log (2 1) ( 1) ( , ) l N i s i s l i i i i i s s t z q a q q q t z z ? ? ? ? = ? ? ? ? ? ? = ? ? ? ? ? ? ? ? ? ? ? ? + + ? ? ? ? ? ? ? ? ? ? ? ? ? ? The updated equation of 1 i q at ( l +1)th iteration is 1 1 2 1 ( ) ( 1) 1 ( ) ( ) ( ) ( ) 0 0 ( , ) 1 (9) log (2 1) ( 1) ( , ) i i i i l N l i s s l l l l i i s s t z q a q q q t z z ? ? ? ? + = ? ? ? ? ? ? = ? ? ? ? ? ? ? ? ? ? ? ? + + ? ? ? ? ? ? ? ? ? ? ? ? ? ? For updating the parameter 2 i q , i = 1, 2, ?, K we consider the derivative of ( )( ) ; l Q ? ? with respect to 2 i q and equate it to zero. We have ( ) ( ) ( ) ; log ( ; ) l l Q E L ? ? ? ? ? ? = ? ? Therefore ( ) ( ) 2 ; 0 l i Q q ? ? ? = ? implies ( ) 2 log ( ; ) 0 l i L E q ? ? ? ? ? = ? ? ? ? ? ( ) ( ) ( ) ( ) ( )1 1 2 ( , ) log ( , ) log 0 K N l l l i s i s i i s i t z f z q ? ? ? = = ? ? ? + = ? ? ? ? ? ?? ( ) 1 2 ( ) 0 1 2 0 2 1 ( , )(10) ( ) log ( 2 1) ( ) ( , ) The mixing parameters ? i and the model parameters q i1 , q i2 are usually considered as known apriori. Drawing a random sample from the entire image ( Mclanchan G. and Peel D. (2000)) is the most commonly used method for initializing parameters. This method shows better performance, if the sample size is large and its computational time is heavily increased. When the sample size is small, there are some small regions which may not be sampled. To divide the whole image into various homogeneous regions we use the K -Means algorithm. In this algorithm the centroids of the clusters are recomputed as soon as the pixel joins a cluster. N l i s s i N l s i i i i i s s i t z q z a q q q t z a ? ? ? ? = = = ? ? ? + ? ? + ? ? ? ? ? ? ? The updated equation of 2 i q at ( l +1)th iteration is 2 1 2 2 ( ) ( 1)1( ) ( ) ( ) ( ) 0 0 1 ( , )(11)( ) log ( 2 1) ( ) ( , ) i i i i N l i s l s N l l l l s i i s s i t z q z a q q q t z a ? ? ? ? + = = = ? ? ? + ? ? + ? ? ? ? ? ? ? Where ( ) ( ) ( ) ( ) ( ) 1 ( , ) ( , ) ( , ) l l l i i s i s K l l i i s i f z t z f z ? ? ? ? ? = = ? IV. # a) K-Means Clustering Algorithm It is one of the simplest clustering technique with a primary goal to find the partition of the data which minimizes the squared error or the sum of squared distances between all the points and their respective cluster centers (Rose H. Turi, ( 2001)). This K-means algorithm uses an iterative procedure and this procedure minimizes the sum of distances from each object to its cluster centroid, over all clusters. This procedure consists of the following steps. 1) Randomly choose K data points from the whole dataset as initial clusters. These data points represent initial cluster centroids. 2) Calculate Euclidean distance of each data point from each cluster centre and assign the data points to its nearest cluster centre. 3) Calculate new cluster centre so that squared error distance of each cluster should be minimum. 4) Repeat step 2 and 3 until clustering centers do not change. 5) Stop the process. In the above algorithm, once only if all points have been allocated to their closed cluster centre then the cluster centers are updated. The advantage of this algorithm is that it is a very simple method, and based on intuition about the nature of a cluster, which is that the within cluster error should be as small as possible. The disadvantage of the K-Means algorithm is that the number of clusters must be supplied as a parameter, leading to the user having to decide what the best number of clusters for the image is (Rose H. Turi, (2001)). Success of K-means algorithm depends on the parameter K, number of clusters in image. After determining the final values of K (number of regions) , we obtain the initial estimates of 1 2 , , and i i i i a q q ? for the i th region using the segmented region pixel intensities with Pearson Type VI distribution .The initial estimate ? i is taken as ? i =1/K, where i = 1,2,...,K. The parameters 1 i q and 2 i q are estimated by the method of moments as first moment µ 1 and its three central moments (µ 2, µ 3 and µ 4 ). # b) Hierarchical Clustering Algorithm In order to utilize the EM algorithm we have to initialize the parameter ? i and the model parameters q i1 , q i2 which are generally considered as known apriori. The initial values of ? i can be taken as ? i =1/K where, K is the number of image regions obtained from the Hierarchical clustering algorithm (Marr D. et al (1980)). The steps involved in hierarchical clustering algorithm are as follows. Step 1: Start by assigning each item to a segment. Each of the N items, are associated with N segments, each containing just one item. Let the distances (similarities) between the segments be the same as the distances (similarities) between the items they contain. Step 2: Find the closest (most similar) pair of segments and merge them into a single segment. The number of segments is now reduced by one. Compute distances (similarities) between the new segments and each of the old segments. Step 3: Repeat steps 2 and 3 until all items are segmented. Step 3 can be done in different ways, namely i) Single-Linkage ii) Complete-Linkage and iii) Average-Linkage segmenting. We consider the Average -Linkage methodology. Average-Linkage segmenting (also called the unweighted pair-group method using arithmetic averages), is one of the most widely used hierarchical clustering algorithms. The average linkage algorithm is obtained by defining the distance between two segments to be the average distance between a point in one segment and a point in the other segment. The algorithm is an agglomerative scheme that erases rows and columns in the proximity matrix as old segments are merged into new ones. The proximity matrix is D = [d(i,j)]. The segments are assigned sequence numbers 0,1.., (n-1) and L(k) is the level of the K th segment. A segment with a sequence number m is denoted by (m) and the proximity between segments (r) and (s) is denoted d [(r), (s)]. The algorithm is composed of the following steps: Begin with the disjoint segment having level L(0) = 0 and sequence number m = 0. Find the average dissimilar pair of segments in the current segment, say pair [(r), (s)],for all pairs of segments in the current segment. and (s) and adding a row and column corresponding to the newly formed segment. The proximity between the new segment, denoted (r, s) and old segment( K )is defined in this way. i j (r,s) K (r,s) K d(i,j) N N d = ? ? where d(i, j) is the distance between object i in the cluster ( r, s ) and object j in the cluster K, and N (r,s) and N (k) are the number of items in the clusters ( r, s ) and K respectively. The above procedure is repeated till the distance between two clusters is less than the specified threshold value. We obtain the initial estimates of q i1 , q i2 and ? i for the i th region using the segmented region pixel intensities with the moment method given by Pearsonian Type VI distribution, only after determining the final values of K (number of regions). After getting these initial estimates, the final refined estimates of the parameters through EM algorithm given in section (III) is obtained. V. # Segmentation Algorithm In this section, the characteristics of the image segmentation algorithm are projected. After refining the parameters, the first step in image segmentation is allocating the pixels to the segments of the image. This operation is performed by Segmentation Algorithm which consists of four steps. Step 1: Plot the histogram of the whole image. Step 2: Obtain the initial estimates of the model parameters using K-Means algorithm and moment estimates for each image region as discussed in section IV. Step 3: Obtain the refined estimates of the model parameters q i1 ,q i2 and ? i for i=1, 2, ..., K using the EM algorithm with the updated equations given by ( 7), ( 9) and ( 11) respectively in section III. Step 4: Assign each pixel into the corresponding j th region (segment) according to the maximum likelihood of the j th component L j. That is , ( ) ( ) 2 1 2 1 j ( 1) 1 2 2 L ( ) ( 1, 1) max j j j j q q s i s q q j k j j j j z a z a q q q ? ? ? + ? ? ? ? = ? ? ? ? + ? ? ? ? 1 2 , . i i j j a z q q ? ? < ? ? < < ? VI. # Experimental Results The performance of the developed a segmentation method for the natural images, which are considered on the earth. For implementing this algorithm, we need to initialize the model parameters, which are usually done by using moment method of estimations. Initially the feature vector is divided into The initial estimates of the number of the regions K in each image are obtained and given in Table 1. Table 1 : Initial Estimates Of K IMAGE TIGER EAGLE NEST BIRD FACE Estimate of K 2 3 4 4 From Table 1, we observe that the image TIGER has two segments, images EAGLE has three segments and images NEST BIRD AND FACE have four segments each. The initial values of the model parameters mi1,mi2, qi1,qi2 and ?i for i = 1, 2, ?, K , for each image region are computed by the method given in section III. By making use of these initial estimates and the updated equations of the EM Algorithm given in Section III, the final estimates of the model parameters for each image are obtained and presented in Tables 2.a,2.b, 2.c, and 2.d for different images. 1) 06892 0.988622 0.00316 7.37 0.0032 i i l s s s z z f z z z ? ? + + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? + ? ? ? ? ? ? ? ? ? = + + ? ? + ? ? 0.0626 1.06892( 1) 1.06892 0.06256 000) ( 1 0.06 6 ,2 1) ? ?s i i z z E ? ? ? + + + ? ? + ? ? ? ? + ? ? ? ? ? ? ? ? ? ? + + 1, .642036 0.405 1 3 ) + + The estimated probability density function of the pixel intensities of the image NEST BIRD is ? + + + + ? ? ? ? + ? ? ? ? ? ? ? ? ? + + + The estimated probability density function of the pixel intensities of the image FACE is ( i i l s z z f z ? ? + ? ? ? ? + ? ? ? ? ? ? ? ? ? = + + + + ( ) ( )(i i z z ? + + ? ? ? ? + ? ? ? ? ? ? ? ? ? + + + Using the estimated probability density function and image segmentation algorithm given in section III, the image segmentation is done for the five images under consideration. The original and segmented images are shown in Figure 2 # Initialization of Parameters by Hierarchical Clustering Algorithm In this section, we evaluate the efficiency of the proposed image segmentation algorithm. For this purpose of evaluation the images are collected from the Berkeley image data set. For this we need to randomly pick four images from the database and feature vector consisting of gray value for each pixel of the image. The feature vector in each image is modeled by using Pearsonian Type VI & Pearsonian Type I Distribution. By dividing all the pixel into different regions using Hierarchical Clustering Algorithm, the initial values of model parameters mi1, mi2, qi1,qi2 and ?i are obtained. Using these values and the updated equations of EM-Algorithm discussed in section III with MATLAB code, the final values of the model parameters are calculated and presented in the tables 3.a, 3.b, 3 ( D D D D ) Year 2014 F Table-3.a Table-3.b Table-3.c Substituting the final estimates of the model parameters, the probability density function of pixel intensities of each image is estimated. The estimated probability density function of the pixel intensities of the image TIGER is ( ) (0.6101) (1.7071) (0.6101)(1.7071) ( ) (0i i l s z z f z ? ? + + ? ? ? ? + ? ? ? ? ? ? ? ? ? = + + + + (40.i i z z ? + + ? ? ? ? + ? ? ? ? ? ? ? ? ? + + + # The estimated probability density function of the pixel intensities the image EAGLE is The estimated probability density function of the pixel intensities of the image NEST BIRD is ( ) (0.31705) (0.7971) (0.31705) (0.7971) ( ) (0.31705 0.7971 1) ( (0.8124)(-0.39174) (-0.8320) 1 1 -0.39174 -0.8320 , (-0.39174 -0.8320) (0.31705 1, 0.7971 1) (0.6205)(-0.2174) i i l s z z f z ? ? + + ? ? ? ? + ? ? ? ? ? ? ? ? ? = + + + + (2.i i i z z z ? + + ? ? ? ? + ? ? ? ? ? ? ? ? ? + + + ? ? + ? ? ? ? + (0.i i i z z z ? + + ? ? ? ? ? ? ? + + + ? ? ? ? + ? ? ? ? ? ? ? ? ? + +65i i i i z z z z ? + + ? ? ? ? + ? ? ? ? ? ? ? ? ? + + + ? ? + ? ? ? ? ? + (0.i i z z ? + + + ? ? ? ? ? ? + + + ? ? ? ? + ? ? ? ? ? ? ? ? ? + + .4979i i i i z z z z ? + + ? ? ? ? + ? ? ? ? ? ? ? ? ? + + + ? ? + ? ? ? ? ? + (0.? ? + + + + ? ? ? ? ? ? + + + ? ? ? ? + ? ? ? ? ? ? ? ? ? + + +1 # Performance Evalution In this paper we have conducted the experiment and also examined its performance by making use of the image segmentation algorithm. The performance evaluation of this segmentation technique is carried by obtaining the three performance measures From Table 4 it (iii) variation of information (VOI). By computing the segmentation performance measures namely VOI, PRI and GCE for the five images under study using Pearsonian Type VI Distribution (PTVID-K), the performance of the developed algorithm is studied. The computed values of the performance measures for the developed algorithm and the earlier existing finite Gaussian mixture model(GMM) with K-means algorithm and Hierarchical algorithm are presented in Table 4 for a comparative study. For the above retrieved images FACE, NEST BIRD AND EAGLE The calculated image quality measures using proposed PTVID and GMM with Kmeans and Hierarchical algorithm are displayed in the Table 5. # Table 5 : Comparative Study of Image Quality Metrics Image Segmentation for Animal Images using Finite Mixture of Pearson type VI Distribution mentioned quality metrics we can retrieve images accurately by using the proposed algorithm. A comparative study is done on the algorithm based on finite Gaussian mixture model with the proposed algorithm and it reveals that the MSE of the proposed model is less than that of the finite Gaussian mixture model. The performance of the proposed model in retrieving the images is better than the finite Gaussian mixture model by making use of these quality metrics. # IX. # Conclusion In this paper, by using finite mixture of Pearsonian Type VI distribution a new model image segmentation is introduced and analyzed. The pixel intensities of animal images better characterizes the mixture of Pearsonian Type VI distribution which is validated through experiment with Berkeley image data set. The model parameters are estimated by using the EM Algorithm. By using the Maximum Likelihood estimates, the Segmentation Algorithm is developed under Bayesian framework. The Experiment on the Berkeley image data set reveals that this image segmentation method outperforms in segmenting the animal images then that of the existing algorithm basing on Gaussian mixture model with respect to image segmentation quality metrics such as PRI, GCE and VOI. The proposed algorithm is much useful for image analysis and image retrievals. The image that is developed can be extended with a K-dimensional feature vector for color images which will be takes as elsewhere. It is perceived that all the image quality measures for the five images are meeting the standard criteria as given in the Table 5. Basing on the above # IMA GE Quality![Initialization of the Parameters by k -Means and Hierarchical Algorithm Generally the efficiency of the EM algorithm depends upon the count of the regions in the image, during the estimation of the parameters. The number of Volume XIV Issue III Version I mixture components taken for K -Means algorithm is, by plotting the histogram of the pixel intensities of the whole image. The number of peaks in the histogram can be taken as the initial value of the number of regions K.](image-2.png "F") 1![Increment the sequence number: m = m +1. Merge segments (r) and (s) into a single segment to form the next segmenting m. Set the level of this segmenting to L(m) = d[ (r), (s) ]. 2. Update the proximity matrix, D, by deleting the rows and columns corresponding to segments (r)](image-3.png "1 .") 1![Figure 1 : Histograms Of The Images](image-4.png "Figure 1 :") 2![Figure 2 : Original and Segmented Images ORIGINAL IMAGES](image-5.png "Figure 2 :") ![, 0.4465 1) +The estimated probability density function of the pixel intensities of the image BIRD isUsing the estimated probability density function and image segmentation algorithm given in section V, the image segmentation is done for the four images under consideration. The original and segmented images are shown in Figure3.](image-6.png "") 3![Figure 3 : Original and Segmented Images ORIGINAL IMAGES](image-7.png "Figure 3 :") ![is identified that the PRI values of the existing algorithm based on finite Gaussian Mixture model the five images considered for experimentation are less than the values from the segmentation algorithm based Pearsonian Type VI Distribution with K-means. Similarly GCE and VOI values of the proposed algorithm are less than that of finite Gaussian mixture model. This reveals the fact that the proposed algorithm outperforms the existing algorithm based on the finite Gaussian mixture model. After developing the image segmentation method , it is required to verify the utility of segmentation in model building of the image for image retrieval. By subjective image quality testing or by objective image quality testing the performance evaluation of the retrieved image can be done. Since the numerical results of an objective measure allow a consistent comparison of different algorithms the objective image quality testing methods are often used. There are several image quality measures available for performance evaluation of the image segmentation method. An extensive survey of quality measures is given by Eskicioglu A.M. and Fisher P.S. (1995). Using the estimated probability density functions of the images under consideration the Image Segmentation for Animal Images using Finite Mixture of Pearson type VI Distribution retrieved images are obtained and are shown in Figure 4. namely, (i) probabilistic rand index (PRI), (ii) global consistence error (GCE) and](image-8.png "") 4![Figure 4 : The Original and Retrieved Images ORIGINAL IMAGES](image-9.png "Figure 4 :") .cand 3.dEstimated Values Of The Parameters For TIGER ImageNumber of Image Regions (K =2)Estimation of InitialEstimation of FinalParametersParameters byParameters by EMHierarchical clusteringAlgorithmImage RegionImage Region1212?i0.5000.5002.01257-1.012571 i a-75.717-11.7563-1.6508-0.0459i a282.970740.67088.3755-28057.911 i m0.47710.01560.610140.2429i m2-0.5228-0.98431.70710.01562Estimated Values Of The Parameters For EAGLE ImageNumber of Image Regions (K =3)ParametersEstimation of Initial Parameters byEstimation of Final Parameters by EMHierarchical clusteringAlgorithmImage RegionImage Region123123?i0.3330.3330.3330.45740.19210.35031 i a-29.580-99.0809-41.2789-0.19680.6095-0.05994i a222.56714.152335.7243-0.7845-0.70016-1.08021 i m0.56720.875010.536060.268312.083062.50880i m2-0.4327-0.1249-0.46390.677420.616150.44170Estimated Values Of The Parameters For NEST BIRD ImageNumber of Image Regions (K =4)Estimation of Initial Parameters byEstimation of Final Parameters by EMParametersHierarchical clusteringAlgorithmImage RegionImage Region12341234?i0.2500.2500.2500.250.81240.6205-0.1252-0.30771 i a-29.742-42.662-22.721-13.326-0.39174-0.2174-0.5579-0.0467i a222.29918.10827.07275.188-0.8320-0.757217.4511018.71 i m0.57150.70200.45630.15060.317052.24802.70255.3865i m2-0.4284-0.2979-0.5436-0.84940.79710.53490.39050.14647 4PERFORMACEIMAGES METHODMEASURESPRIGCEVOIGMM0.8234 0.4956 2.568TIGERPTVID-K0.9896 0.4742 1.921PTVID-H0.9897 0.4762 1.920GMM0.8423 0.7006 8.354EAGLEPTVID-K0.8505 0.7109 7.577PTVID-H0.8627 0.7054 7.2002 © 2014 Global Journals Inc. (US) Global Journal of Computer Science and Technology © 2014 Global Journals Inc. (US) * Color Image Segmentation: Advances and Prospects" Pattern Recognition Cheng 2001 34 * Image Quality Measures and their Performance MAEskicioglu PSFisher IEEE Transactions On Communications 43 12 1995 * Image Segmentation and Retrievals based on Finite Doubly Truncated Bivariate Gaussian Mixture Model and K-Means KGvs Rajkumar Srinivasa AndPRao -Srinivasa Rao International Journal of Computer Applications (IJCA) 25 4 2011 Accepted for Publication * A Practical Hand Book on Image segmentation for Scientific Applications Jahne 1995 CRC Press * Statistical approach to Xray CT imaging and its applications in image analysis PAKelly IEEE Trans. Med. Imag 11 1 1998 * Performance Evaluation of Finite Normal Mixture Model -Based Image Segmentation Techniques TLei IEEE Transactions On Image Processing 12 10 2003 * A survey of genentic algorithms applications for image enhancement and segmentation MantasPaulinas AndriusUsinskas Information Technology and control 36 3 2007 * Theory of Edge Detection DMarr Proceedings of Royal Society London 207 1980 * Image Segmentation Based on a Finite Generalized New Symmetric Mixture Model with K -Means MSeshashayee KRao ChSatyanarayana And PSrinivasaRao International journal of Computer Science Issues 8 3 2011 * MSeshashayee KRao ChSatyanarayana And PSrinivasaRao Studies on Image Segmentation method Based on a New Symmetric Mixture Model with K -Means 2011 11 * Image Segmentation for Animal Images using Finite Mixture of Pearson type VI Distribution conference on Adv. Computer control * Unsupervised image segmentation using generalized Gaussian distribution with hierarchical clustering KSrinivas Yerramalle PV G DRao -Prasad Reddy Journal of advanced research in computer engineering 4 1 2010 * Unsupervised image classification using finite truncated Gaussian mixture model KSrinivasaSrinivas Yerramalle And Rao Journal of Ultra Science for Physical Sciences 19 1 2007 * Cluster Based Image Segmentation HRose Turi 2001 Monash University, Australia Ph.d Thesis * GMclanchan DPeel The EM Algorithm For Parameter Estimations New York John Wiley and Sons 2000 * Comparing Clustering -An axiomatic view Meila proceedings of the 22nd International Conference on Machine Learning the 22nd International Conference on Machine Learning 2005 * Variational learning for Gaussian Mixtures NNasios AGBors IEEE Transactions on Systems, Man and Cybernetics 36 4 2006 Part B : Cybernetics * A Review On Image Segmentation Techniques SKPal NRPal Pattern Recognition 26 9 1993 * Srinivasa Rao And Srinivas supervised image segmentation using finite Generalized Gaussian mixture model with EM & K-Means algorithm PV G DPrasad Reddy K International Journal of Computer Science and Network Security 7 4 * Srinivasa Rao And Srinivas Yerramalle-(2007), supervised image segmentation using finite Generalized Gaussian mixture model with EM & K-Means algorithm PV G DPrasad Reddy K International Journal of Computer Science and Network Security 7 4 * Unsupervised Image Segmentation based on Finite Doubly Truncated Gaussian Mixture model with K-Means algorithm YSrinivas International Journal of Physical Sciences 19 2007 * Image segmentation-A State-Of-Art survey for Prediction ShitalRaut 2009 International