Research On Character Recognition Technology In License Plate Recognition System

The license plate character recognition technology is one of the key technologies in the Intelligent Transportation Systems, which involves many fields such as pattern recognition, artificial intelligence, computer vision and digital image processing. Many scholars at home and abroad specialize in the license plate location and license plate character recognition technology, and achieved some results, but there are many worthy of continued research and improvement. For example, now the accuracy rate of most of the license plate recognition method for high-definition image is high, but in the case of weather changes, uneven illumination, as well as license plates exist stains, deformation and tilt, the recognition rate dropped significantly, so raising all-weather robustness of the license plate recognition technology makes sense for us.License plate recognition is divided into three steps:license plate location, character segmentation and character recognition. Character recognition as the last part of the license plate recognition system, is also the most important link, and the choice of recognition algorithm directly impact the accuracy of the recognition rate of the whole system. At present, the most commonly used methods of license plate character recognition have template matching, neural networks and support vector machines. However, due to the restrictions of many objective conditions, the recognition effect using a single classification algorithm is often not ideal. This paper analyzes and summarizes the characteristics of commonly used character recognition algorithms and their defects on the license plate character recognition application. we propose a two-stage classification architecture on license plate character recognition, which combines Support Vector Machine (SVM) with character local feature extraction.The input of our algorithm is the license plate character after pretreatment. With this method, first, rough classification is performed, using Kernel Principal Component Analysis (KPCA) as feature extraction algorithm on license plate images, and SVM is used to classify. If the result of rough classification is not easily confused characters, further sub-classification is not needed. If the result of rough classification is easily confused characters, the next fine classifier is necessary. For the different characters with different local feature, different classification methods are designed to distinguish them, and the final recognition results can be obtained.The experimental results show that compared with single classification algorithm the, presented combination classification method has higher accuracy for license plate characters with complex backgrounds, and greatly reduce the time-consuming.