Research On The Restoration And Recognition Algorithm Of Motion Blurred Vehicle License Plate Image

The restoration of motion blurred image is an important subject of imageprocessing. It has a great effect on the further application of image. The restorationof motion blurred image has been widely applied in various fields, such as the trafficmonitoring system. With the development of economy, the number of the mobilevehicles is increasing rapidly, which has brought many difficult problems to thetraffic control. In the traffic monitoring system, due to the high-speed motion, theimage of the license plate is often blurred by the motion. It brings great difficulty tothe follow-up image analysis and target recognition, and it brings greatinconvenience to traffic management. In this paper, the main research object is themotion blurred license plate image, and the main work is as follows:(1) This paper mainly introduced the principles of inverse filtering, wienerfiltering, and Lucy-Richardson algorithm. On the basis of experimental results, theadvantages and disadvantages of three methods were analyzed. In order to eliminatethe ringing effect in the restored image, the mirror extension had been put forward inthe paper. The simulation experiment results show that the mirror extension has anobvious effect on suppressing ringing effect.(2) The point spread function of motion blurred image is the key to the restoration.An accurate point spread function can make the restored image much closer to theoriginal image. However, the point spread function of actual image is unknown, so itneeds to be accurately estimated. The paper puts forward a method to estimate the motion blurred parameters. According to the spectrum characteristics of linear motionblurred image, this paper puts forward a method that using Hough transform to detectmotion blurred direction, and motion blurred length is estimated by the differentialautocorrelation function curve. Simulation experiments of artificial fuzzy motionblurred image have been done by MATLAB, and the experiment results show that themethod of motion blur parameter estimation is high accuracy and small error. Theparameters of actual motion blurred image are acquired by the method. According tothe parameters, the actual image had been restored by wiener filtering recovery method,and the restored image effect is good.(3) In order to the further application of license plate image, the license platelocation, license plate character segmentation and license plate character recognitionare studied in this paper. The license plate is located by the method of mathematicalmorphology. The license plate image projection curve and the character characteristicare applied in the license plate character segmentation. Finally, the template matchingmethod is used to recognize the character of license plate. In order to match thetemplate, the feature of Chinese character stroke and image’s energy value is extracted.While, the feature of letter and number in the license plate image is extracted, such asleft and right profiles and energy value. Simulation results indicate that the method hasgood performance.