Feature Analysis And Recognition Of Altered License Plate Image

The topic of this thesis is the research on the feature analysis and recognition of the altered license plate image."Alteration" refers to the alteration of the true contents of relevant documents and articles by means of alteration,excavation,splicing and other means by the non-modification right holder,so as to distort the truth and pretend to be the true contents of the original original.Alter the image of the license plate by means of magnets and other objects to attach fake license plate Numbers and letters to the original image of the license plate,so as to avoid the violation of the law.The paper recognizes the altered license plate image by image processing technology.The recognition of altered license plate image consists of three main parts:image preprocessing,image feature extraction and classification recognition.In view of the pre-processing part of the altered license plate image,several image denoising and image enhancement algorithms are analyzed and compared.According to the requirements of the altered license plate image recognition,the detailed features of the image need to be retained as far as possible.The image is denoised by bilateral filtering,and the image is enhanced by histogram enhancement algorithm.For the alteration of license plate image feature extraction part,the global and local features of the image are studied respectively.On the extraction of color features,according to this characteristic,the color saturation between the tamper part of the altered license plate image and other regions is different,convert the image to HSV color space,the gray level difference of the histogram and three color moments under the s-channel of the image are extracted as the color features of the image.At the same time of color feature extraction,the boundary histogram,SIFT feature descriptor and color and boundary mixed features of the sample license plate image were extracted,and the recognition accuracy was comprehensively compared through different feature data.For the identification part of the altered license plate,three pattern recognition algorithms including BP neural network,Euclidean distance and support vector machine are analyzed,considering the small number of samples in this experiment and the characteristics of dichotomy,SVM was selected as the feature classifier.By combining each feature data with different kernel functions,the recognition accuracy of different kernel functions is verified,and the experimental results are summarized.