作　　者： (陶彬娇）; (陈倩）; (潘中良）; (李萍）;

机构地区： 华南师范大学物理与电信工程学院,广州510006

出　　处： 《激光杂志》 2017年第8期97-101,共5页

摘　　要： 分块加权颜色直方图算法应用于服装检索时效果并不理想。针对存在的问题,对传统算法进行改进。首先采用3×3不均匀分块方式,通过大量实验获取了最佳分块大小和加权权重比。其次,结合Grabcut图像分割算法,提出了一种基于Grabcut图像分割的前景色分块加权颜色直方图算法。对纯色、带少量图案和花色这三类服装,分简单和复杂背景两种情况,分别进行实验。实验结果表明,相比两种传统算法,改进算法很好的解决了服装背景色干扰的问题。对于三类服装的检索,改进算法的查准率和查全率均比另两种传统算法有了显著提高。尤其是对于复杂背景的情况,改进算法的优势更加明显。 The effect of block weighted color histogram algorithm applied to clothing retrieval is not ideal. This paper studies the existing problems of the traditional algorithms and improves them. First of all,by segmenting every image into uneven blocks of 3＊ 3,the optimal block size and the weighted weight ratio are obtained from many experiments. Then,a new algorithm of foreground color block weighted color histogram based on Grabcut image segmentation algorithm is proposed. Experiments are respectively done on three categories of clothing： pure color,with a small amount of pattern,and color. They are divided into two cases： simple background and complicated background. The results show that,compared with the two traditional algorithms,the improved algorithm solves the problem of the interference of clothing background color very well. And the precision ratio and recall ratio of the improved algorithm have been improved significantly for three categories of clothing,especially in the case of complicated background,the improved algorithm has obvious advantages.

关 键 词： 服装图像检索 分块加权颜色直方图算法 图像分割算法