National costume image retrieval based on integrated region matching

Aiming at the shortcomings of traditional image retrieval based on global features, this paper presents an effective images retrieval method of national costume based on integrated regional matching, in consideration of the significant visual features of national costume images. Firstly, the image is divided into multi-regions by fuzzy color, and a multi-feature fusion is formed to extract color, texture and shape feature for each region. Then, due to the differences of regional importance, a reasonable weight is proposed fusing three different regional importance factors, including saliency, area percentage and color difference. Finally, the integrated regional matching algorithm is used to retrieve the national costume images. In this method, the fuzzy color region division validly avoids the inaccurate segmentation caused by the simple segmentation algorithm and high computational complexity caused by the complex one. At the same time, the introduction of the regional weight factors well inhibits the negative effect produced by the complex background and the interference region, so that the robustness of the algorithm is improved. The experimental results show that the algorithm proposed can express the visual features of national costume images more fully and accurately, and achieve good retrieval performance. It will provide a valid means of processing technology for the national costume images in relevant area.