【摘要】 本文主要研究服装检索系统中颜色特征提取算法。本文首先分析了传统颜色直方图算法用于服装检索时存在的几个问题,接着分别采用切割图像直方图算法、颜色矩算法和前景色颜色直方图算法进行服装图像的颜色特征提取。本文对三类服装:纯色、带少量图案和花色服装分别进行了实验,实验结果表明,三种算法各有不同的适用范围和优缺点。在此基础上,本文将上述三种算法进行线性加权组合,提出了一种基于加权欧氏距离的改进算法。测试过程中,通过修改三种算法的权重比调整各算法的权重,并对三类服装分别计算其查准率和查全率,分析不同权重比下改进算法的检索效果。大量实验结果表明,改进算法对于三类服装的检索效果都比较稳定。特别是对于带少量图案的服装,改进算法比单一算法的查准率有显著提高。实验还发现,当切割图像颜色直方图、颜色矩和前景色颜色直方图算法的权重比为1∶4∶5时,检索效果最好,无论是查准率还是查全率都较高,均优于单一算法。更多还原

【Abstract】 This thesis mainly researches the color feature extraction algorithms in the clothing image retrieval system.Firstly it analyses several problems about clothing image retrieval while using traditional color histogram algorithms. Then it extracts the color feature of clothing images by using cutting image color histogram algorithm,color moment algorithm and foreground color histogram algorithm respectively. Three categories of clothing- pure color,with a small amount of pattern,design and color- are tested. The result shows that three algorithms have different applicable scope and their own merits and drawbacks. On that basis,it combines the aforementioned three algorithms with linear weighting and proposes an improved algorithm based on weighting Euclidean metric. In the process of test,it analyzes the improved algorithm’s retrieval effects by changing the weight of the three algorithms to calculate the precision ratio and the recall ratio for these three categories of clothing respectively. The results show that the improved algorithm is relatively stable for each category of clothing. Especially for the clothing with a small amount of pattern,the precision ratio of improved algorithm is significant increased,compared to other single algorithm. Also it shows that the retrieval effect is the best while the weight ratio of cutting image color histogram algorithm,color moment algorithm and foreground color histogram algorithm is 1 ∶ 4 ∶ 5.Both the precision ratio and the recall ratio are much higher than other single algorithm.更多还原