Research Of Image Reconstruction Algorithms Based On Block Compressed Sensing

At present, the license plate recognition system has been widely used in highways’ toll and speed monitoring management, peccancy monitoring of urban traffic intersection, parking management and many other areas, it becomes an important part of intelligent transportation system.This thesis mainly researches motion blur license plate image restoration and locating, tilt correction,character segmentation and character recognition on the basis of analying the researches in recent years, and the relevant algorithms are improved.First of all, on the basis of analying the existing motion blur image restoration algorithm, a motion blur image restoration algorithm based on PSF estimation and adaptive TV coupling gradient fidelity term is proposed. The experimental results show that the proposed method can well realize the recovery of the motion blur vehicle image.In license plate locating stage, considering that the image quality after recovery still has certain gap with the ideal image, which is likely to affect the accuracy of the traditional positioning methods,this thesis implements a license plate location method based on chaos adaptive genetic algorithm and projection method. The experimental results show that the proposed method still has good positioning effect under the condition of poor image quality.In tilt correction stage, the Radon transformation method is used for horizontal tilt correction,while rotation projection method is used for vertical tilt correction. The results show that it reduces the tilt correction time while does not affect the precision.In character segmentation stage, we first conduct character coarse segmentation using the projection method, then use the method of connecting markers and growth to further character segmentation until all of the characters are segmented.In character recognition stage, this thesis uses the LS-SVM method based on wavelet kernel function,and analyzes the feature extraction and classifier design.Finally the algorithm in this thesis and the traditional character recognition algorithms are compared through the experiment.