Product Image Retrieval Based On MPEG-7 Feature Fusion

With the rapid development of Internet and the multimedia technology, images play an important role in e-commerce as a part of products increasingly. How to seek out the product image which user wish quickly and accurately in massive multimedia information is becoming an urgent problem need to be solved. Content-based image retrieval is expected to be the key technology to solve the above problem. Due to the description method of image feature lack of criterion, the universal property of Content-based image retrieval system is poor. MPEG-7 has solved the problem of image description standard. Single image underlying feature reflects the properties of image from different angles, so effective feature fusion can improve retrieval performance.This dissertation studies product image retrieval based on feature fusion from the deep and weighted fusion respectively. In the deep fusion retrieval, Color quantization histogram, Color layout and Angular radial statistic, information of feature contains the meaning of deep fusion itself; these three features are extracted and used in image retrieval respectively. Retrieval results show that each feature appears good image retrieval performance according to different characteristics of product image. In the weighted fusion retrieval, on the basis of analysis of image retrieval based on the above three deep fusion feature, this dissertation put forward a kind of image retrieval method based on color and shape feature fusion of MPEG-7.First of all, Color quantization histogram, Color layout and Angular radial statistic is normalized inside respectively; then each feature's similarity is calculated; then each feature is normalized outside respectively; finally, the proportion value to the three features is distributed objectively according to the three retrieval performance value, the fusion feature is gotten. Test the image retrieval based on the fusion feature to 10 kinds of the product image database.After a mass of testing and comparison, the results show that the retrieval method based on MPEG-7 color and shape feature fusion devised in this treatise has better retrieval performance compared with the retrieval based on single feature. The advantages of each feature are assembled. The results provide a powerful basis to product image retrieval based on MPEG-7 feature fusion in e-commerce.