Research On Recognition Mechanisms For Deformation Text Based On Shape Context

In recent years, method based on deformable template matching information recognition has become an active research direction in computer graph model processing. The most important content in this field is how to extract features for description the recognition object, and how to develop better approach to get recognition results.In this dissertation, we focus on recognition mechanisms and feature extraction methods for deformation text images and trademark image that both have deformation text and graph. The research is mainly about deformation text image recognition and deformation trademark image recognition. The contributions of our work include the following five parts:(1) The image model database has been set up for our research, including the two parts:text image database and car trademark model database; (2) Several mainstream feature extraction methods be analyzed carefully in theory which are belonged to deformation text recognition system, and the advantages and disadvantages of various methods be evaluated, then make a series of tests based on search results in the MATLAB platform, and through the performance comparison of the recognition experimental results to find appropriate feature extraction method for the deformation text recognition;(3) A new similarity metric distance mechanism has be proposed which combined by some single distance measure method by the theory of fuzzy inference, to improve the single view of the current recognition system similarity metric distance method. This method has many recognition advantages of high recognition accuracy and strong robustness, and has strong application value in the deformation text recognition field; (4)In the basis of the study by (2) and (3), we found the appropriate feature extraction method for deformation image, the feature extraction algorithm based on Shape context was proposed, and the new similarity distance mechanism mixed the theory of fuzzy inference, and thus built the deformation text image recognition system. We carried out validation experiments by the above (1) database. The recognition results show that this system compared with the traditional algorithm system has certain advantages in recognition performance, and the strong robustness; (5) Extended the deformation text recognition algorithm, we established the new recognition system for deformation trademark images in which both have text and graph. We carried out a lot of validation experiments for this system, and analyzed the experimental results. The experimental results show that our algorithm and system is effective and has high recognition accuracy.