A Study On Techniques Of License Plate Recognition Based On Mathematical Morphology And Support Vector Machine

The License Plate Recognition (LPR) system is a significant sector of the modern intelligent traffic management, which has a far-reaching future for the application of traffic management and police security, and attracted by all over the world at the present.Four key parts of license plate recognition system have been researched, which are license plate location, license plate edge detection, character segmentation and character recognition.I. License plate locationAiming at the problem of multi-area in mathematical morphology method, a license plate localization method based on mathematical morphology and color features has been proposed. License plate region contains rich edge informations, so the image edges are extracted out, and then the image can be smoothed with morphological method, after that candidate regions are ascertained by Binary-Area-Morphology, and the region of real plate is filtrated finally by color features of plate. Experimental results show that such method is not only accurate and robust, but also practical.II. license plate edge detectionAiming at the familiar problem of noise disturbance, an image edge detection method of multi-structure elements and multi-scale based on mathematical morphology is proposed. The proper structure elements could be obtained by controlling the threshold and numbers of accumulation. The algorithm can effectively remove the noise while detecting the edge details with multi-direction structure elements.III.Character segmentationAiming at the fuzzy license plate image whose histogram is not clearly dual-peak distributed, an image binarization algorithm based on information entropy and iterative method is presented. Then the characters are segmented with the an improved vertical projection method. Estimate the initial point by vertical projection and prior knowledge. The results of simulation demonstrate that the method gained favorable results.IV. Character recognition Aiming at the problem of sample position deviation in traditional SVM classification algorithm, a position normalization algorithm based on centroid and outside edge of Characters has been adopted. This method improves the effectiveness of feature extraction. On the basis of two-class classification, a multi-class classifier that are suitable for the feature of license plate is constructed. And a optimization Algorithm of parameter setting is proposed to improve the recognition rate.