Research Of Image Retrieval Technology Based On Clothing Elements

Nowadays, online shopping is becoming more and more popular and clothing online shopping has become a trend. However, the diversity of clothing, such as clothing style,fabric style, the price of clothing, the brand of clothing and so on, makes purchasers spend a lot of time on choosing right clothes. The clothing shopping technology based on text retrieval can't meet the needs of consumers. Therefore, image retrieval technology based on CBIR (Content-based image retrieval) occupies a very important position in the clothing retrieval and a rapid development opportunity.The object of this thesis is the technology of image retrieval system based on clothing elements. The three basic elements of clothing is clothing fabrics, clothing styles and clothing color. The popular search keywords on the shopping sites such as: leather dress, tassel dress, fur collar coat, pleated skirt and so on, which also belong to the retrieval of the three basic elements. Based on the basic elements of clothing and the situation of website retrieval, this thesis mainly focuses on the identification of garment fabric and the recognition of clothing style. The main research work and achievements are as follows:1. An improved Tamura algorithm and a fabric recognition algorithm based on decision tree are proposed. According to the fabric texture, weaving method, color and other characteristics, we use coarseness, directionality, contrast and color feature description of fabric to build a decision tree model, which achieved the purpose of classification of apparel fabrics.2. A clothing style recognition algorithm is proposed based on HOG and SVM.Because of the complexity of the clothing style, the interference of the background and the highlight of the local elements, we cut the picture into the ROI region and choose two kinds of ROI regions: collars, other clothing elements. Through the training of a large number of features in those ROI regions, two classifiers trained by SVM can recognize the ROI area of the garment style. The experimental results show that the classification algorithm proposed in this paper has good performance and the recognition rate is high.3. An image retrieval system is developed. The system is based on clothing fabric and clothing style recognition algorithm proposed in this paper. The system uses MySQL as the database and realizes the development of a B/S architecture website based on Servlet and JSP Technology. The experimental results show that the algorithm proposed in this paper can promote the performance of clothing retrieval.