Adaptive content based image retrieval based on RICE algorithm selection model

In this paper, we propose a framework for “Algorithm Selection” for image retrieval by content (CBIR). The framework is based on the model of RICE and is adapted to satisfy a given query depending on its characteristics by choosing the best classical CBIR-Algorithm from an Algorithm-Portfolio. As many as six algorithms for content based image retrieval have been included in the framework as alternatives for the different queries, including the training step. These algorithms range from RGB color moments, RGB color histogram to local binary pattern (LBP), etc. Therefore, there has been put an effort in the framework to cover the basic characteristics of images: Color and texture. Also, the framework integrates two color models to better enhance the Algorithm-Query adaptation process. Experimentations on the Wang (Corel 1k) database show the effectiveness of the proposed framework. Indeed, enhancements of more than 4% in precision have been obtained.