利用相关反馈技术和形状特征的服装图像检索算法

Research on image retrieval algorithm by relevance feedback technology and shape features

基于内容的图像检索在服装电子商务中是受关注的研究热点。为了提高基于形状特征的服装图像检索准确性,提出一种利用通用Fourier描述子与改进的相关反馈技术以提高服装图像检索精度的算法。分析比较了各种形式的Fourier描述子,选用性能突出的通用Fourier描述子,按照系数能量比率优化了距离计算公式,同时根据样本的概率分布改进了查询向量优化算法中的权值计算方式。相较于传统算法,以裙装为例的实验结果证明所提出算法的优势,并且在结合纹理、色彩等特征后,算法性能得到进一步的提高。

Content-based image retrieval is a research focus in clothing electronic commerce. In order to improve the accuracy of the clothing image retrieval system based on shape feature, an algorithm has been put forward, which uses general Fourier descriptors and the improved feedback technology. By analysis of various kinds of Fourier descriptors, the general Fourier descriptors which have outstanding performance are selected and distance computation formula is optimized by energy ratio of coefficient. According to the probability distribution, the algorithm improves the weight calculation method for the vector optimization algorithm. The experimental results of skirts show that the proposed algorithm outperforms the conventional algorithm. By combining such features as color, texture, etc. , the performance of the proposed method can be further improved.