Image Segmentation for Animal Images using Finite Mixture of Pearson Type VI Distribution
Image Segmentation is one of the significant tool for analyzing images, the feature vector of the images are different for different types of images. In remote sensing, Environmental ecological systems, forest studies, conservation of rare animals, the animal images are more important. In this paper we developed and analyze an image segmentation algorithm using mixture of Pearson Type VI Distribution. The Pearsonian Type VI Distribution will characterize the image regions of animal images. The appropriateness Pearsonian Type VI distribution for the pixel intensities of image region in animal images is carried by fitting Pearsonian Type VI Distribution to set of animal images taken from Berkeley image data set. The image segmentation algorithm is developed using EM algorithm for estimating the parameters of the model and maximum likelihood for image component under Bayesian framework. For fast convergence of EM algorithm the initial estimates of the model parameters are obtained by dividing the whole image into K image regions using K-means and Hierarchical clustering algorithm and utilizing the moment method of estimates. The performance of proposed algorithm is studied by conducting an experiment with set of animal images and computing image quality metrics such as PRI, GCE and VOI. A comparative study of developed image segmentation by Gaussian Mixture model and found the proposed algorithm performed better for animal images due to asymmetrically distributed nature of pixel intensities in the image regions.
