Probabilistic model based image retrieval using hypothesis testing

Abstract

The content based image retrieval (CBIR) is aimed to find the most similar images from a collection of images or a database to the query image according to the visual or semantic similarity. Current image retrieval algorithms mainly focus on feature selection and learning distance metric. Other methods are also proposed to promote the performance such as relevance feedback. But as a whole their performance still can not reach the expectation for application. This paper presents a new framework for image retrieval by introducing the hypothesis testing heory. The rest of this paper is organized as follows: Section II introduces the motivation of the proposed approach and the hypothesis testing problem in image retrieval. Section III is the main part of the paper and in this section the concepts of background model, image model, test ratio and feature extraction that compose the framework are introduced and discussed. Section IV shows the relevant experimental results of the algorithm. © 2012 Springer-Verlag London Limited.