Research On Cognizing Character Extraction And Recognition Of Video Image Character

Scientific Research is always made on to broaden the faculty of human being, which focused on the aspect of physical strength previously, while the ability to feel and perceive nowadays after the invention of computers. For this, there have come forth many new subjects as well as Computer Vision. Consequently, projects of Image Segmentation, Extraction of Region of Interested and Character Recognition come into being.Relating to bionics, the image is transformed in order to segment and extract the region including char conveniently. Considering the former methods of character exaction, we give prominence to the stroke location, run length and amount feature of character, and recognize the character based on it.The following are the major contents in the thesis.1. We provided a new method that tried to transform the whole image to featured space and facilitated identification and self-elimination of disturbance according to the hue and luminosity features of eyes of human being. We gave the definition and description of the transformed space. A wavelet-texture filter was designed to distill space- transform- based feature of the image itself was executed.2. We raised a new model that we disassemble the character into several parts, which could be recognized by computer topologically based on the high-frequency wavelet coefficients vector, disregarding the traditional extraction method that used the statistical or structural feature based on the individual pixel in the 2-dim plane of character. Moreover, the concept of multi-dim cognizing feature model was put forward by encoding the character, according to its' location and run-length information. The information confusion and redundancy could be largely eliminated, as leaded to the improving of the preciseness when recognizing the character.3. We extracted the Multi-dim-cognizing feature based on the application object of VLP char, and we developed arithmetic of character extraction and recognition based on large number of samples and met with good results.