特征提取与匹配算法在民族服饰图案上的应用研究  

Research on Application of Feature Extraction and Matching Algorithms in Chinese Costume Patterns

作　　者：陈金广[1] 曹春梅[1] 任冰青[1] 马丽丽[1] 

机构地区：[1]西安工程大学计算机科学学院,陕西西安710048

出　　处：《丝绸》2015年第5期36-41,共6页Journal of Silk

基　　金：国家自然科学基金项目(61201118);"好运来创新研发基金"资助项目(HYL201405)

摘　　要：服饰图案包含某些象征意义,是传统服饰文化研究中的重要元素。将图像处理技术应用于民族服饰图案的特征提取时,尺度不变特征变换(SIFT)算法和快速鲁棒性尺度不变(SURF)算法是两类较典型的特征提取算法,它们在图像旋转、尺度变化、噪声干扰情况下具有较好的适应性。文章选取清代宫廷服饰中具有代表性的图案作为实验对象,用SIFT和SURF算法分别对该图像进行特征提取,用最优节点优先(BBF)算法确定特征点匹配点对,计算两种算法的正确匹配率和时间复杂度。实验结果表明,两种算法在服饰图案发生旋转、尺度变化、噪声干扰情况下均保持较高的特征匹配率,对从事网络中海量服饰图像数据自动化分类和检索方面的研究人员具有一定参考价值。Dress patterns contain some symbolic meanings and are important elements in the research on traditional dress culture. This paper applies image processing technology to feature extraction of folk costume patterns. SIFT and SURF algorithms are two typical feature extraction algorithms and have good adaptability under the condition of image rotation,scale change and noise jamming. With a representative pattern in palace dress in Qing dynasty as experimental object,this paper conducts feature extraction of this image respectively with SIFT and SURF algorithms,determines matching point pair of feature points with best-bin-first( BBF) algorithm and calculates correct matching rate and time complexity of the two algorithms. The experimental result shows that both algorithms have high feature matching rate under the condition of dress image rotation,scale change and additive noise jamming. This conclusion has certain reference value for researchers engaged in automated classification and retrieval of mass garment image data online.

关 键 词：服饰图案 SIFT SURF 特征提取 图像处理 民族服饰 图案分类 

分 类 号：TS941.19[轻工技术与工程—服装设计与工程]