基于连衣裙结构特征匹配的款式识别技术

Style recognition technique based on feature matching in dress construction

为便于人们快速检索相似款式的服装,以连衣裙款式关键点轮廓为基础,对轮廓的不同描述和相似性度量计算的检索方法进行了研究。将形状上下文、传统欧式距离和基于系数分配的欧式距离3种算法计算的相似度结果与人工测评的相似度测评表做对比,确定最优算法。结果表明:形状上下文匹配方法的效果最佳,其中准确度和精确度分别为82.91%和77%,可作为轮廓相似性度量算法;基于系数分配的欧式距离与传统欧式距离的匹配效果相比,准确度和精确度都有所提升,分别提高6.45%和0.44%;根据计算初步推断,在轮廓特征描述子中按照相似度影响程度大小排序为:角度余弦值、宽度、高度。

In order to facilitate rapid retrieval of similar styles of clothing,this paper reports a study on different descriptions of dress contour and on the retrieval methods for measuring and calculating dress similarity,based mainly on the key contour point of dresses.The similarity results were calculated by the three algorithms,i.e.,the shape context,traditional Euclidean distance and Euclidean distance based on coefficient assignment,and the results were compared to identify the optimal algorithm.The results show that the shape context matching method has the best effect,and the accuracy and precision are 82.91%and 77%,respectively,which can be used as the contour similarity measure algorithm.The accuracy and precision of the Euclidean distance based on the coefficient allocation are improved by 6.45%and 0.44%respectively.Based on calculation,it is preliminarily concluded that the degree of similarity influence,represented by a contour feature descriptor,is ranked in the order of angle cosine value,width and height.