The Research On Content And Emotion Based Product Image Hierarchical Retrieval

With the rapid development of the computer and communication, network changes a lot. As a new business model, e-commerce is growing very rapidly. However, compared to traditional shopping, there are considerable disadvantages for e-commerce, one of which lacks interaction between users and goods. Because users do not know enough goods in the online, many users can not make up its mind shopping. Because the display of goods in e-commerce is mostly in the form of pictures, content-based image retrieval and emotion-based image retrieval can provide a new search method to meet the needs of the users' search.To solve the above problem, the paper studies content-based image retrieval and emotion-based image retrieval . Firstly, the paper describes the background, significance and status of international research. Then, content-based product image retrieval is discussed in detail. Visual feature extraction, similarity measurement and relevance feedback are key technologies for content-based image retrieval systems, which are important parts of a complete and effective image retrieval system. Color, texture and shape of the image are three important visual features, product image retrieval based on color, texture and shape is introduced in depth.Secondly, the paper presents a product image retrieval method based on the improved SIFT matching algorithm. SIFT can effectively describe the local image features, and it's robust to rotation, illumination, scale, affine and noise transformation. Matching the number of key points is only taken into account in the traditional SIFT matching algorithm, but the distance measure between the matched keypoints also influences the similarity measure, and the shorter the distance between the matched keypoints, the lower the similarity measure. Given the above ideas, the paper presents a novel similarity match method, which uses the number of the matched keypoints and the distance between the matched keypoints to calculate the similarity measure from global image and local sub-image, and using the SIFT matching algorithm to retrieval product image.Next, the paper describes the emotion-based product image retrieval. The algorithm contains the indexing process, the query process and the relevance feedback. Once an emotion query is selected from the preceding emotion adjectives, the associated query color and query gray codes are selected, and these codes are compared with every DB code in the images database to obtain the top N database images most similar to the query emotion. Also, a new relevance feedback and a weight update algorithm are proposed that automatically update the inter-code weights between the query color code and query gray code and intra-code weights of the query color code.Finally, fusing product image content and emotion, the paper presents a hierarchical framework for product image retrieval, which has two stages of rough match and exact match. Rough match stage uses product image emotion information, and exact match stage uses product image visual features. The approach generally meets the users' search request and it's proved to be effective through the different experiments. The research on content and emotion based product image hierarchical retrieval have theory significance and specific applications, and it provide a new solution for interaction between users and goods.