Content-based Clothing Image Retrieval System

Online shopping develops fast in recent years,the most popular goods is the clothing products. There are several reasons to explain the development, for example, commodity prices are lower than in the shop, and there is more choice for customer buying one clothes. Besides, people can wait at home, and later your cloth will be sub-delivered to your door. Such advantages attract more and more people shopping online. Buy clothing products online rely on the visual information's retrieval of the image; this is the nature of this kind of sales model. The mainstream shopping site still use keyword-based image retrieval technology, it is contrary to the nature of the online shopping for clothing, and it is unable to retrieve clothing accurately to meet people's requirements, its bottleneck due to its own development can not meet the e-commerce development. The CBIR technology applied to the goods on behalf of a new type of search engine development, there are some shopping sites abroad using CBIR technology to achieve the visual shopping search.This paper designed and implemented a content-based clothing image retrieval system. The system included display modules, processing module, feature extraction module, similarity measure modu. Besides, analyzes module processes to achieve the function, the image are crawled from taobao.com and like.com. CCIRS use MATLAB 6.0 as the tools,it provides a simple user-friendly interface.Image database calssified in storage and provied a test set for algorithm.Experiments show CCIRS is operating normally,the functions are relized.It is a practical platform for feature extraction algorithm.This paper do deeply research of the thesis of the related systems and visual shopping site's technology, by comparing the features of the natural landscape images, product images, and apparel products image, it summed up the unique features of clothing product image, for example, the object is prominent, fixed spatial information, color variety, pattern are rich, discrimination relay on the detail. According to the visual characteristics of the clothing image, this paper selected and implemented five kinds of feature extraction algorithms for the color, shape and texture of clothing which are modified histogram, cumulative modified histogram, edge direction histogram, wavelet modulus maxima 7 moment, Tamura. The system of this paper test the five algorithms, uses the precision of the search result to evaluated and take this as the basis to compare the performance of different algorithms. Correction color histogram, revised color histogram cumulative good search results,edge direction histogram highlight the search results in the clothing patterns image database, wavelet modulus maxima 7 moment which values the outer contour has a good search result, Tamura is robust. The findings have a positive significance on the application of CBIR in the visual shopping.