Research Of Online Trademark Image Retrieval Based On Content

Whether a trademark can be registered is determined by comparing it with those registeredtrademarks. The current retrieval methods are mostly manual, which cost greatly and low efficiency.Therefore, it is necessary to establish an efficient trademark retrieval system. However, the currentresearches on content-based trademark image retrieval technology are mostly based on globalimage features, and often ignore the effects of local features on the whole image visual effect, orhave some deficiencies in retrieval precision and efficiency, scope of application and theconsistency of retrieval results with human’s visual feelings. So it is highly important to establishan accurate and efficient trademark retrieval system.The trademark image retrieval technologies based on content are introduced in this paper first,and the current research status is analyzed. Then, the key technologies of trademark image retrievalare described in detail, especially for the Scale Invariant Feature Transform method, which exhibitsthe advantages and disadvantages by experimental analysis. Thus, according to the defect of SIFT,a new algorithm of feature extraction and matching based on the average moment of feature pointsis proposed. It extracts local features by dividing images into several sub-blocks, and has invarianceof rotation, scale and translation. The experiment shows that the features of image blocks in thisalgorithm are not sensitive to noise and subtle variations of image edges, and have a favorableretrieval performance. Meanwhile, the relevance feedback of retrieval results makes the retrievalsystem meet customer requirements better.In the last part of the paper, an online trademark retrieval system is designed, which is basedon the model of Browser/Server. It includes two design modules, i.e. feature extraction and buildingof feature database, online trademark retrieval. This system verifies that the algorithm in this paperis feasible, and makes further research and an attempt of apply of the universal online trademarkretrieval system.