摘要
基于传统的人工定义的图像检索,预设信息模糊,检索效率低且准确性差,难以满足日益增多的服饰图案素材库的检索需求.本文从服饰图案的纹理和形状出发,基于尺度不变特征(SIFT)向量变换技术,提取服饰图案的局部特征点及其描述子,并通过归一化互相关(NCC)算法实现图像的相似度判定.检索系统基于后台和实时分段查找的机制,对服饰图像素材库进行快速识别和定位.实验表明:该方法能够实现图案的准确定位和相似性判定,且对视角变化、仿射变换、噪声等具有良好的鲁棒性.图像检索系统的实现免除了繁冗的文字信息定义,提高了图像信息的识别度,对面料图案设计、服装设计、电子商务图像快速搜索等具有重要的意义.
Image retrieval based on traditional artificial definition shows low efficiency and poor accuracy for ambiguous text message, which cannot meet the retrieval of increasing dress pattern library. This paper presents an image retrieval method that uses modified scale invariant feature transfor- mation （SIFT） and normalized cross correlation （NCC） to characterize and identify the similar images based on textures and shapes. Besides, two stages of background-real feature screening scheme can effectively improve images detection speed. Experiments show that the above method which shows good robustness to perspective change, affine transformation, noise, etc. can quickly achieve accurate pattern position. The retrieval system can eliminate abundance text messages defined, and be helpful to fabric pattern design, fashion design and e-commerce image retrieval.
出处
《北京服装学院学报（自然科学版）》
CAS
2013年第4期43-49,共7页
Journal of Beijing Institute of Fashion Technology：Natural Science Edition
基金
中央高校基本科研业务费专项资金(CUSF-DH-D-2014063)
