Research And Application On The Content-based Clothing Image Retrieval System

With the popularity of the Internet and the rapid development of e-commerce technology, online shopping is more and more affecting people’s spending and shopping habits. Clothes as the most popular online shopping products are becoming the highest turnover of goods, with its characteristics of reasonable prices, selection, and delivery.The mainstream online shopping websites such as Taobao, Jingdong and Dangdang, retrieved clothing through the keywords entered by users to describe the clothes. In this process, consumers need to rely on their visual perception of the clothes and then express in term of text language on clothing. User’s intervention for the clothing image description is not accurate, and users need to spend a lot of time on the clothing to make the selection. In order to improve search experience, many Internet companies put into image search on clothing studies using the visual characteristics. But depending upon image low-level features describing the image will cause the missing of semantic. Overall, in this paper, we use clothing image online shopping site to realize the clothing image mapping keywords through feature extraction, classification and image coding. Finally, we complete the research and application on the content-based clothing image retrieval system by keyword-based search. The main works are as follows:(1) The traditional color feature extraction algorithm is analyzed through researching the related technologies of content-based image retrieval. The paper presents color feature extraction algorithm improvement over traditional methods. Using the main color analysis, color space domain codebook is obtained. And then encoding image color feature in clothes according to the codebook, statistical color feature is obtained. Finally the processing of statistic characteristic is to get a lower dimension vector based on color feature attributes location.(2) About the feature classification algorithm in the design of clothing, different from the classification algorithm based on fusion feature, the paper proposed a decision tree classification training method for clothing image of color, texture and shape feature, respectively. The decision tree acquired through training is to code image feature, realizing feature coding to keywords mapping. Experiments show that the proposed decision tree has practical value for image classification methods to achieve the keyword mapping.(3) This paper is a consolidation of image feature extraction, feature coding and feature and keyword mapping module, based on key Clothing retrieval technology, designs and implements the clothing image retrieval system based on content. The system is a semantic-based image retrieval system realizing clothing retrieval by the way combining mapping the user input image to keywords and keyword-based search.