Research Of License Plate Recognition System Based On Wavelet Packet And Support Vector Machine

In order to achieve accurate automatic license plate recognition, the paper proposes the use of wavelet packet and support vector machine method, which has strong anti- interference, can adapt to a wide range of application of various scenarios.The main contents of this paper are:License plate positioning portion, the first image of the license plate pretreatment, pretreatment methods include graying, image smoothing, image enhancement, and other commonly used methods of image processing, the role of this part is to remove interference noise, strengthen the image features. Regional positioning plate is divided into two parts, the first part of the texture features by positioning the vehicle first strike by the edge detection image edge fea tures, and then use a priori knowledge of morphological operations and draw the target area; a second portion is positioned through color features, first convert an image from RGB space to HIS space, then through the target color H, S, threshold I determined the three components of screening to find the license plate area. Finally, multi- feature integrated obtain the license plate area.Character segmentation section, the license plate was subjected to the first image tilt correction, level correction using least squares fitting method is based on vertical correction is used in Radon transform. Then were seeking correction plate image horizontal projection and vertical projection, and projection curve de-noising, according to a priori knowledge and curve cha racteristics of the character region is divided, and finally get a single character image.License plate character recognition section, the first character feature extraction are discussed here to get the character image feature by optimal wavelet packet decomposition. Thesis using least squares support vector machine license plate character recognition to classify, design and implement Chinese characters, letters, numbers, letters classifier, classification method many, classifier parameters determined by cross-validation method. Finally, all the algorithm simulation, simulation tools platform for matlab. Training and test sets 1000 as a vehicle containing the image library, and have achieved good recognition results.