Segmentation of Natural Images With K-Means and Hierarchical Algorithm Based on Mixture of Pearson Distributions

Image analysis and retrieval are most important for video processing, remote sensing, computer vision and security and surveillance. In an image, regional variation is more important for authentication and identification. Nowadays, the model based segmentation methods are more prominent and provide accurate results for segmentation of images. For ascribing a suitable model it is reasonable to consider the probability model, which closely matches with the physical features of the image region. In this paper a new and novel approach of image segmentation is carried using the Type III Pearsonian system of distributions. In the current work, it is considered that the entire picture is characterized by a K-component combination of Pearsonian Type III distribution. Using the EM algorithm, the performance parameters for the currently considered model are estimated. Through experimentation 4 real images randomly selected from the Berkeley image database. The computed values of PRI, GCE and VOI revealed that proposed method provide more accurate results to the same images in which the image regions are left skewed and having long upper tiles. Through image quality metrics the performance of image retrieval with the proposed method is also studied and found that this method outperforms then that of the segmentation method based on GMM. The results of the current model is being compared with the other existing previous models like the 3-paprameter regression models and the k-means hierarchical clustering models for various sets of input images and the results are displayed in the performance evaluation models chapter in detail.