Research and Implementation on Content-Based Clothing Image Retrieval
|School||Harbin Institute of Technology|
|Course||Computer Science and Technology|
|Keywords||content-based image retrieval the background noise removal the color histogram LBP algorithm|
In the area of electronic commerce, on-line shops often display the images of their goods to attract the customers. Potential consumers in the hope that a comprehensive understanding of the commodity before to decide whether or not to purchase, and what information consumers can access, including commodities description on the text and the tagged images of commodities. However, text information is not intuitive, consumers often need images of goods, but images of goods usually are tagged not accurately by the business men, which make it difficult for consumers to get information about the product. Content-based image retrieval can avoid these problems and help consumers efficiently access to these images of related products.A content based image retrieval system for electronic commerce clothing image is proposed in this paper. Through statistical analysis, we found that these clothing images have the following characteristics: first, each image there is only one region in which users may be interested. Second, the background color of the images is complex: it is often the general clutter of real scenes.To deal with the characteristics of clothing images, a content-based image retrieval system is proposed, which is based on the technique of background-noise removal. First, in order to extract the outline of clothing in the images, a garment-outline extraction method based on the JSEG segmentation algorithm is proposed in the paper. The algorithm applies the approximation method on both sides of images and the linear interpolation method to get the outline of clothing. Second, this paper pays much attention on the methods of content-based image retrieval systems, especially on the feature extraction technology of color and texture. Color histogram is used in the system, and its performance is very well due to the removal of background-noise. The retrieval results of experiment show that color histogram in HSV space are better than that in RGB space. LBP algorithm (Local Binary Pattern) is used to extract the texture feature, through the experiment on the clothing images of stripe texture, it proves that the LBP algorithm is suitable to extract the texture feature of clothing. For the combination of color and texture feature, the experimental results achieved the expected goals. Finally, a user-friendly image retrieval system is implemented, which supports user-defined retrieval mode, that is, based on color or texture or based on both of the two features.