Research On License Plate Recognition Technology Based On Blurred Image

License plate automatic identification system uses computer to analyze the vehicle image and identify the license plate number information,as an essential link in transportation management has a particularly wide range of applications.It plays a very important role in intelligent traffic management and information technology.At present,the research on the automatic identification technology of the license plate has achieved good results,and the correct recognition rate of the existing system can reach more than 90%.However,in some special cases,such as the relatively high speed movement between the image acquisition device and the vehicle,a clear image of the vehicle can not be grasped,resulting in the phenomenon of motion blur in the vehicle image.This phenomenon has a serious negative impact on the license plate recognition so that the correct recognition rate is reduced.This paper focuses on the study of the license plate recognition method with motion blur phenomenon and makes appropriate improvements.The main components are recovery of motion blurred images,restoration image preprocessing and license plate location,license plate image binarization,character segmentation and character recognition.The first step of this paper is to restore the motion blurred vehicle image.Motion blurred images can be seen as the result of convolution of the original clear image and a degenerate function,so the first job to restore it is to estimate the degradation function.Only by accurately estimating the parameters of the degenerate function(fuzzy angle and fuzzy length)can we establish a more accurate degradation function,and then choose the appropriate recovery method to restore the fuzzy image processing.When the fuzzy angle is estimated,the motion spectrum is simulated by Fourier transform,and then use the Hough transform to detect the fuzzy direction of the line in the spectral image,so as to determine the fuzzy direction angle.The method used to estimate the fuzzy length is to rotate the motion blurred image in the horizontal direction according to the motion blur angle,then the fuzzy autocorrelation function is used to calculate the fuzzy length.This leads to two necessary parameters of the degenerate function.Finally,the motion blurred image is reconstructed by Wiener filter according to the obtained degradation function.The second step is to locate the restored image of the license plate.It is necessary to pre-process the fuzzy restored image before positioning,which includes image contrast enhancement and smoothing filtering and other operations.The aim is to improve the quality of motion blurred images to increase the accuracy of positioning.In the license plate positioning,the Roberts operator edge detection is used for the image based on the richness of the edge information of the license plate area,and then through a series of expansion operations,corrosion operations and open and close operations and other mathematical morphology to accurately locate the location of the license plate.The license plate of the determined position is divided on the fuzzy restoration image.After obtaining a separate license plate image,the OTSU threshold algorithm is binarized.And then remove unnecessary borders and small dots and other useless information.The following is the license plate image segmentation and normalization,the character segmentation takes the vertical projection plus the character width threshold limit method to accurately separate the individual characters from each other and normalize them for identification.Finally,the normalized characters are to be recognized.In this paper,the template matching method and BP neural network method are used to identify and compare the results.The template matching method uses the improved eigenvalue extraction method to compare the standard template character image with the character image to be recognized in turn,the highest degree of matching is the corresponding recognition results;BP neural network identification method must first build a good network of a large number of sample training,until the neural network training mature to meet the application requirements.The comparison of the results of the two recognition methods shows that the BP neural network identification method is stronger than the template matching method,and the correct recognition rate is higher.At the end of this paper,we use the collected sample images to test and verify the results of each step,and then analyze and summarize the results,put forward the shortcomings of this study and the prospect of future research.