Regional Shape Description Method And Its Application To Trademark Image Retrieval

The research of trademark retrieval and classification has great practical significance for trademark registration, safeguarding the trademark owner’s interests. Traditional trademark classification is by written records and markings. This method isn’t very efficient, and makes much error, but costs time and labor, requires high experience and skills. A fast and efficient image retrieval and classification technology is needed to recognize the exponential growth of trademarks. Content-Based Image Retrieval (CBIR) is a hot spot of intelligent information technology, and this provides a viable solution to image retrieval and classification.This paper studies the area shape description method, which extracts the effective area characteristics of the target shape, and applied to trademark image retrieval. The main work is as follows:(1) Research on the description method of shape regional characters, which is based on Adaptive Hierarchical Density Histogram (AHDH), and focuses on feature extraction of area shape, feature normalization and combinations of features, etc. Then we analysis the time complexity of AHDH and its problems in detail.(2) Based on the research AHDH, we proposed a new regional shape feature descriptor-Rotating Hierarchical Density (RHD). When the image is rotated, we select sub-region according to rotation of the image segmentation criterion, then do sub-dividing of sub-region and select it’s sub-divided region, extract pixel characteristics and envelope rectangular area of sub-regions in each layer by rotation and segmentation, we choose rotation angle from 0° to 360° by uniform sampling, so that the feature descriptor keeps a rotation, scaling, translation invariance in describing the region shape.(3) Design retrieval scheme, test and compare the retrieval property by using AHDH, Zernike Moments and RHD to retrieve the collected MPEG-7 Image Library, which includes 1400 images of 70 different categories. Then use RHD to identify and retrieval the trademark image library TradeMark70.Experiment shows that the RHD method has some improvements compared with AHDH, which keeps translation, scaling and rotation invariant, but also avoids high computational complexity and storage overflow problem when its high iterative segmentation level. RHD has greatly reduced the time complexity of retrieval.