Triplet-Based Regularized Diffusion Process for Improving Visual Retrieval

Abstract

As the object retrieval problem cannot be well solved by pairwise distances, many algorithms have been developed for improving visual retrieval result. In recent studies, it is demonstrated that the contextual (dis-)similarities based on diffusion process can be obtained directly by solving an optimization problem with a general regularization framework, which contains a smoothness constraint and a fitting constraint. To improve the effectiveness of visual retrieval, this paper introduces a novel smoothness constraint named triplet-based smoothness constraint into the regularization framework of diffusion process. According to the random walk model, the proposed model can be used to simultaneously regularize three elements, and provides a quite different form of high-order information. Then we proposed the triplet-based regularized diffusion process by introducing the hybrid fitting constraint into our method. The experimental results on different shape databases demonstrate that retrieval results can be effectively improved by using the proposed methods.