Video Character Recognition Technology Research And Application

Video character information is very important in many fields like labeling and retrievaling. Now it is unguency for us to extract video character information. In this thesis we discuss video recognition system which includes image pretreatment, text zone location, text tracking, text binary, character segmentation, character recognition and so on.In character segmentation, we give a series of concepts such as character gap width sequence, character width sequence, character gap variance, character gap deviation, character width deviation and character composedness. We can segment the different characters accurately.In character tracking, we propose the concepts of marginal density, observing window and similarity. Using these concepts we design the algorithm of back-searching and forward-searching the ending frame where the aiming character locates on in video frames.In character recognition, we take the projection of image in the directions that are 0°,45°,90°,135°and circle on as its feature. The circle projection needs to be calculated by the wavelet decomposition and fractal. We employ multi classifier which is composed of K-means clustering and multi BP neural net work. Finally, we give full and detailed experiments according to the algorithms proposed in this thesis. The experimental results turn out these algorithms have high efficiency. Thus we develop an applicable video Chinese character recognition system.