Research On Video-based Character Recognition

Character Recognition has attracted more attention as an important branch of pattern recognition. Along with the rapid growth of information of images and videos, how to use the character information included in images and videos to establish content-based indexing of image information has been a very meaningful task.Video is consisted of frames of images. In order to recognize the characters included in the video information, firstly, we should preprocess the video files according to the relative knowledge of video and image processing. Such treatments includes extraction of key frames from video by the method of DirectShow, change colorful images to gray images, gray balance and deletion of noise and so on.On the basis of preprocessing, an algorithm with aim to differentiate between Chinese and English characters and to divide the adhesive segment. Segment lines of text firstly, then divide every line of text into individual words for the further processing such as feature extraction and character recognition according to the known features of character. This algorithm distinguished Chinese and English according to their different characteristics, then do different processing on Chinese and English separately.After the division of character, we use BP algorithm which belongs to neural network to recognize character. As the traditional BP algorithm is inadequate, an improved BP algorithm is given here. It improves BP mainly through restructuring gradient in the SIGMOID. If the value is too large, the output (0 or 1) of all floors may be discrete, and the results deteriorated. In the other hand, if the value is to small, the linear capability become strong and nonlinear capability weakened, so the best value is between those two ones. To form new activation function by change the improved gradient into new one, and the gradient of SIGMOID function can be adjusted to the optimum value during the training.Many experiment results show substantial that recognition accuracy rate of this system is satisfied.