Using Shape Features For Trademark Image Retrieval

With the increase of register trademark, there will consume many time and effort for human to retrieve and compare a trademark with large trademark database. There are many limitations in current trademark retrieval method, including time consuming, inaccuracy, low automation degree. Aiming at those shortages, we study and proposed a trademark retrieval method based on shape features, and optimized the index structure of image database.On the basis of generic Content-based Image Retrieval theory and considering trademark image features, we proposed a new trademark image retrieval method based on shape features. The proposed modified Zernike Moment descriptor is obtained by first normalizing the original shape, to be invariant to planar shape distortions, dividing it into two parts of inner and outer regions with a radius, which is fixed by modificatory k-mean clustering, and calculating the Zernike moment of the parts separately. Weighted Euclidean distance is used for computing distance measure between two shapes.As to the problem of query optimization, aiming at the high-dimensional vector space of trademark image database, we present a new efficient index structure, called Depth Distance Index Structure (DDIS), for k-Nearest Neighbor (KNN) search. DDIS makes pyramid partitioning for the data set, and selects a reference centre respectively, then transforms each data into a single dimensional space associate with pyramid depth and distance to reference centre which belonged to, that are suit to be indexed using a B+-tree structure.The experiments based on the Da Meng Image Retrieval (DMIR) system and gray trademark image database show that, the accuracy of the new proposed retrieval algorithms is exceeding over 36% than invariant moment, and over 11% than original Zernike moment; the response time of KNN search of the new proposed index structure is 70% less than iDistance algorithms, and 50% less than pyramid algorithms, moreover it could return accurate results in small search radius.