Robust Algorithms For Character Recognition

The license plate recognition (LPR) is one of the key technologies in the intelligent traffic system, including the location, segmentation of license plate and character recognition. Although, there are some commercial OCR software products, they are not effective in license plate character recognition, because the image quality of license plate is usually poor. So it is necessary to develop robust algorithms of license plate recognition. In recent years, the character image recognition is applied to counterfeit detector, recognition of banknote serial number is essential for high-end counterfeit detector.Based on the study of the character recognition method, we implement several character recognition algorithms, propose appropriate scheme for the recognition of banknotes character and license plate character according to the experimental results. The details are as follows:1. The algorithms of feature extraction and classification.Implement several algorithms of feature extraction, including directional line element feature, uniform grid feature, gradient feature, elastic mesh directional decomposition feature and LBP feature; According to the comparison and analysis of the experimental results, we choose the uniform grid feature to describe letter and number characters of the license plate and banknote images. We also propose a modified LBP feature for Chinese character recognition of license plate. The algorithms of template matching and Modified Quadratic Discriminant Functions (MQDF) classifier are also implemented.2. The feasible scheme for the recognition of license plate characters and banknotes characters.Propose and realize the recognition method for license plate letters, numbers character and banknotes characters using uniform grid feature and the support vector machine(SVM) classifier. For the license plate characters, the precision is98.2%and the recognition time is3.23ms per character with PC. For the banknotes characters, the precision is99.9%and the recognition time is2.84ms per character with PC. Propose and realize the recognition method for chinese character of license plate based on the modified uniform LBP feature and the template matching, the precision is99.2%, the recognition time is1.97ms per character with PC.3. Build the vehicle license plate recognition systemIntegrated the plate localization algorithm, character segmentation algorithm of laboratory with character recognition algorithm of this thesis, we built a complete license plate recognition system including license plate recognition function and database management function.