Types Of Special License Plate Location And Recognition Algorithm Research

With the development of China economic, It is Increasing pressure on road transport, people more and more pay attention to Intelligent Traffic Management System to the research. As the core of intelligent traffic management system, license plate detection and identification becomes still important. Although there are some license plate recognition system related products appear, but the study of the algorithm has not stopped, there are still many scholars in making further study and improvement.Firstly, the paper to in depth study for the present situation and the existing technology of the license plate recognition. The vehicle image for license plate image tilt for the extraction and recognition,then overlooking vehicle image for license plate image extraction and recognition, the complex background of the vehicle image for the license plate image extraction and recognition, in this based on the use of MATLAB to developed a license plate recognition system, which the Ministry of the software Points including the license plate location, license plate license plate character segmentation and character recognition of three modules.Detailed contents of this paper and results are as follows:First, the vehicle image using gray-scale transformation, smoothing the image enhancement, through the analysis of traditional methods, combined with the characteristics of wavelet multi-scale to realize a high-frequency non-linear wavelet-based image enhancement method of the vehicle image enhancement method, combined with edge detection, morphology, projection and so the license plate location algorithm.secondly, on the license plate image binarization, the application of the Radon transform plate tilt correction, and processing methods based on mathematical morphology image denoising plate.In the license plate character segmentation process, the first image under the license plate scanning its structure to remove the border, according to license plate of the vertical projection and based on the method of combining prior knowledge of the license plate character segmentation Finally, the character size normalized, and then use Hilditch thinning algorithm to refine the characters of the license plate of each character with a statistical method to extract the character of the horizontal, vertical, and the overall statistical characteristics. According to BP neural network can be well on network connection weights modified repeatedly to achieve the advantages of the requirements set in order to improve recognition accuracy. Identification when using the three statistical features based on BP neural network for character recognition, and then identify the three networks were the result of threshold processing, the system implementation process, after taking a large number of experiments the threshold of 0.9, if greater than or equal to 0.9, the bit to take "1", otherwise take "0", very good to improve recognition accuracy. Then identify the results of the three bitwise comparison, there are two or more identical, to take the place of the recognition results, in order to achieve license plate character recognition.