A multi-view fusion method for image retrieval

Abstract

Color histogram is an important technique for color image database indexing and retrieving. However, existing color based retrieval techniques are mainly designed for only extracting global or local feature, which cannot provide effective retrieval of images. In this paper, we propose a novel multi-view fusion method for image retrieval by combining the global color with salient regions color feature, which highlights the important characteristics of the salient regions without losing the background information. Firstly, HSV color histogram is quantified rationally as a global descriptor. Secondly, a salient region detection method is introduced to separate the salient regions and the background regions. After that, color histogram of the salient regions is applied to constitute a region-based descriptor. Finally, a CBIR system is designed by using an adaptive weighting method to combine these two descriptors. The relevant retrieval experiments on Corel-1000 show that the proposed approach brings better visual feeling than single feature retrieval, which exceeds at least 9%.