Research On Shape-based Image Retrieval And Trademark Retrieval

Multimedia technology is one of the fastest growing and most active technologies in the field of information technology, multimedia technology includes: voice?images?text?video and communication information. Also with the rapid popularization of the Internet, content based image retrieval has become a research hotspot. Image information mainly include low level visual features, such as color? texture?shape and their combinations. Shape is one of the main reference characteristics, and feature information carried by shape is more consistent with the human visual feature. Although there are many shape feature algorithm, retrieval efficiency needs to be improved. For different applications, appropriate improved shape algorithm should be selected according to the demand of image retrieval system.Firstly, how to effectively describe and represent shapes of image in shape-based image retrieval is the focus of study, and determines the speed and accuracy of matching algorithm. The traditional shape matching algorithm can't completely describe the global and local information of image. Shape-based image retrieval can be improved by assigning different weights to the global and local feature descriptors. In order to simplify global contour feature descriptor, a combination of improved Hough Transform and Zernike moment invariants has been presented. Experiments show that the proposed method has significantly improved on the matching speed and accuracy.Secondly, the improved shape-based retrieval algorithm is applied to existing retrieval system. By studying the existing algorithm of shape-based trademark retrieval system, this paper's focus is on the trademark segmentation method. In order to reduce the workload of manual trademark segmentation, methods of segmentation based on region growing?segmentation based on horizontal projection?segmentation based on vertical projection?threshold segmentation and manual segmentation have been chosen for trademark preprocessing. By experiment tests, appropriate methods have been chosen to process different types of trademark according to the characteristics of trademark and the segmentation algorithm.Finally, a shape-based trademark retrieval system has been developed. By a large number of experiments, most of the trademark can be processed automatically; thus the improved segmentation algorithm is proved to be feasible. Also, the combination of improved Hough Transform and Zernike moment invariants method is proved to have a better retrieval performance. A shape-based trademark system is developed in the developing environment of Visual C++6.0. As an applied frame, this system is used to demonstrate the effectiveness and feasibility of the proposed method. Experimental results show that the algorithm reduces the trademark retrieval time?achieves a good performance and has good application value.