Research On Trademark Image Retrieval Technology Based On Shape Matching

With the development of market economy, the quantity of trademark increases progressively year by year. The tradition trademark image retrieval based on classified or text labeling has many problems, including the very load of manual illustration work, subjectivity, inaccuracy, etc. Content-based image retrieval (CBIR) technology can overcome these problems, so it obtained the very widespread application in the trademark retrieval domain.The methods of CBIR retrieve the images using the characteristics of themselves, such as color, shape, texture and the space position relations. Trademark image as an artificial image, compared with other features, shape is the most remarkable, so we always use the shape feature to distinguish various trademarks This dissertation focuses on the key technologies of the shape matching based trademark image retrieval, including: trademark image segmentation technology, boundary-based shape description method, region-based shape description method, shape feature fusion and matching, sub-image multi-characteristic fusion retrieval, etc. The main contributions of this dissertation can be summarized as follows:1. Several key technologies of trademark image segmentation are deeply investigated and analyzed. A new multi-layer fusion image segmentation method is proposed, named WG-OLR. This method can carry on the automatic image segmentation with high efficiency.2. Research on the boundary-based shape description method of the trademark image. A new shape boundary matching method based on comer detection and Delaunay Triangulation Net, named DT-MATCH, is proposed. This method can carry on fast feature description to the non-complex shape and has the good retrieval efficiency.3. Research on the region-based shape description method of the trademark image. We mainly aim at one kind of sub-area statistics shape description method to carry on quite studies and find out the most suitable shape description method based on the sub-area statistics description thought.4. After analyzing the advantages and disadvantages of the boundary-based and region-based shape description method, in view of the trademark shape, we propose a new total-graph trademark image retrieval method that fuses the boundary and the region features, named BR-MATCH. This method not only has the good match effect, simultaneously has the quick retrieval speed.5. Using the established image segmentation technology, synthesizing to the research on shape description method, simultaneously fusing the color, the space position relations and other characteristics, we propose a new sub-graph trademark image retrieval method based on the multi-characteristic fusion, named SBR-MATCH. Compares with the total-graph retrieval method, the precision of sub-graph retrieval method has the further enhancement.