Neighboring Pixels Based on a Log-linearized Gaussian Mixture Model for Image Segmentation

Abstract

An advanced probabilistic algorithm developed based on log-linearized Gaussian mixture model aims to estimate posteriori probability of neighboring pixel method in image segmentation. We firstly apply the log-linearized Gaussian mixture to develop and determine the mixture and the mixture component of the Gaussian mixture model. Then, the posterior probabilities of each pixel are also identified by using neighboring pixel method. Secondly, employing maximum likelihood technique to simulate the statistic model under our algorithm framework aims to improve accuracy of segmented images and to reduce impacts of noise during image segmentation process. Our research results present good segmentation yields, and the segmented images are more accuracy comparing to the segmented images which obtained by other segmentation methods.