Application Of Pattern Recognition Based On Digital Image In License Recognition

Digital image recognition is one of the most important research directions of pattern recognition. As a core technologies in the Intelligent Transportation System,license plate recognition based on image is widely used in the Intelligent Transportation Management.The theory of pattern classification and its application in the license plate recognition system is studied in this paper. Considering feature extraction and the construction of classifier, which are the two basic problems of pattern classification, the method of designing classifier in specific applications is studied. Based on the theories of pattern classification such as support vector machine (SVM) that efficiently solve the small sample learning problem, and with the help of image processing technology, license segmentation, character extraction and character recognition, which are the three main parts of license plate recognition is studied systemically, and improvements of the solutions are presented.For the license segmentation problem, a multi-stage SVM classifier with texture feature extraction is designed. First, after a discussion of features that may be useful for the classification, texture features including features based on histogram, co-occurrence matrix and edge are chosen to form the feature vector. Furthermore, in order to improve the accuracy and efficiency, a two-stage classifier is designed and each of which use different features as the input.For the character extraction problem, a statistical method of modeling the image is applied because effective feature for training is hard to be extracted. Under the Gaussian Hidden Markov Random Field, the maximum a posteriori (MAP) method is used for classification. Based on the model and the classification method, an enhancement of the classification is achieved by taking multiple frames from the video and embedding the additional information into the classification method.For the character recognition problem, a multi-class SVM classifier with feature extraction is designed. The projection feature of character is used to form the feature space for number and English letter and a new binary scale multi-class SVM classifier is presented. Compared with traditional multi-class methods, the new one decrease the number of two-class classifiers that are needed, and speed up the recognition.The results of the simulation experiments shows that the algorithms presented are effective in improving the accuracy and speed.