Sift-based Trademark Image Retrieval Technology

CBIR (Content-based image retrieval) technology is a research hot spot at present. Compared to the traditional text retrieval, the indexing results are more in line with human perception. Retrieval process including feature extraction, feature comparison, the final output of the retrieval results. Therefore, the image feature extraction and description directly affects the image retrieval accuracy and speed. The trademarks play the more and more important role in the current society and convey abundant information. This thesis analyzed the trademark image matching and retrieval problems, and proposed two solutions.Firstly, the thesis analyzed the present situation and problems of recent research in CBIR, and discussed the approaches to feature point extraction. Secondly, two algorithms of descriptor generation were discussed and researched on, especially on Scale Invariant Feature Transform (SIFT) features. Based on this research, regarding the characteristics of trademark image and the SIFT features, this thesis proposed a new method of simplified SIFT feature algorithm for trademark images retrieval to eliminate all the un-robust and unnecessary feature key points and improve the computing speed, Then constructed an trademark image retrieval system, which used the new algorithm, the comparison experiments show that the method has higher precision and faster speed, and do well in trademark image retrieval system. Third, this thesis proposed a new method based on the SIFT algorithm and the edge detection based on maximum of wavelet transform modulus for trademark images registration. By comparing the experimental results show that this method is more stable and effective.