Probabilistic Similarity Measures Analysis for Remote Sensing Image Retrieval

In image database retrieval there are many classical similarity measures that can be used to find the target image, these measures mostly belong to geometry model from the point of view of the data model, while little attention has been devoted to the studies on methods based on probability density distribution. In this paper we experimentally investigate some probabilistic similarity measures, present two methods for design of the similarity function of two mixture Gaussian distributions, on the basis of the nearest neighbor rule and K nearest neighbor rule respectively. An experimental study was conducted to examine and evaluate the measures for application to image databases, and the experimental results show that the methods based on K nearest neighbor rule achieve better performance