Using Local Binary Patterns and Gaussian Mixture Models to Bridge the Semantic Gap in Content-Based Image Retrieval

Abstract

Content-Based Image Retrieval (CBIR) engines are systems aiming at using the visual features of images in order to find their relevant. Despite the significant efforts that have been made by researchers to develop CBIR systems, they still suffer from the semantic gap between low level image features and high level user concepts. In this paper, we propose a fully automatic learning-based method to bridge this gap. Our method uses a Gaussian Mixture Model (GMM) as a visual model for each concept, where each component within it group images having the same visual appearance. Our method presents a multitude of advantages: 1) allows user to naturally express their needs using a textual query; 2) permit to retrieve images from unlabeled collections using a textual query; 3) It is fully automatic, as it doesn’t require any human intervention. Experimental results show the efficiency of our method and a high accuracy in retrieval has been achieved.