Commodity Search Based On Sparse Representation Of Visual Dictionary

Commodity retrieval is an important application of image retrieval. Traditional commodity retrieval is based on keyword search, which needs a lot of work on Image annotation. What's more, the process of traditional commodity retrieval ignores some other information but keywords, which may cause some problems like information insufficient or low accuracy of commodity description information. In order to solve these problems, the retrieval based on content of image (such as texture, color, shape and so on) has been proposed. To a certain extent, it solves the limitation of text retrieval. With the facts that the shopping websites are offering massive commodity images, Search images by image is increasingly becoming a hot topic in the field of computer intelligence.The paper studies the feature extraction, image classification and feature fusion of image retrieval. Also I experiment it in the self-built commodity database. The contribution are as follows:(1) The paper briefly analyzes the background, significance and the application prospect of commodity retrieval, and then makes a detailed description of the feature extraction and feature matching based on BoW (Bag-of-Visual-Words).(2) Study the sparse representation theory, introduce sparse representation into the framework of image retrieval. Design an image classification method based on sparse representation. Combining visual dictionary with sparse representation, by reconstructing residual minimum, Category the query image roughly and narrow search range.(3) Study the technology of feature fusion. Due to the different feature express different image information, we extract Sift feature, Gist feature, convolutional neural network (CNN) features and HSV color feature of query image, compare retrieval results of different feature and use adaptive the feature fusion method to obtain better search results.(4) Build our own product search database, and every item is marked in the database. Using matlab to build commodity retrieval system based on visual dictionary sparse representation.