Vehicle Tire Traces Of Image Enhancement And Recognition Method

With the explosive increment of vehicles, the criminal cases of escaping after committing traffic offences and utilizing vehicles as criminal tool are on the rise in our country, which bring severe challenges to the social stability and development. The tire traces of vehicles can provide important clues and evidences for the detection of these criminal cases, therefore it is necessary to preserve the tire traces taken from crime scenes of the accidents in the form of digital image on the spots in order to obtain the most actual material evidences from tire traces for further analysis. However, owing to the influences of weather, image acquisition device, human and other factors, there are a large number of circumstances covering the image information during the acquisition and transmission of tire trace images, such as noise, low or high contrast, vagueness and so on. Therefore tire trace images collected from the accidental spots should be processed by some enhancement algorithms and improve the image quality before further analysis, so as to provide more information about accidental vehicles for relevant investigation departments for accurately solving cases. Furthermore, the various tread patterns lead to the diversity of tire traces, which enlarges the bound of searching accidental vehicles and increases the searching difficulty. For the purpose of reducing the searching range of accidental vehicles, helping relevant investigation departments to solve the cases quickly and precisely, and further boosting the stable development of the society, it is essential to classify the tire trace images.This thesis researches the enhancement and classification methods of tread pattern and tire trace images of different qualities, and the main research works are as follows:(1)This thesis does several researches relevant to the important role of tire trace images in the criminal cases of escaping after committing traffic offences and using the vehicles as criminal tool, and summarizes the international and national research situations as well as the problems existing in current researches about tire trace images.(2)According to the theoretical model and distribution characters in tire trace images of salt and pepper noise, this thesis presents a salt and pepper noise filtering method based on decision analysis and adaptive median filter algorithm. The approach classifies the pixels of tire trace images contaminated by salt and pepper noises by means of the idea of decision analysis, and adopts the corresponding disposal ways towards different classification circumstances. Experimental results demonstrate that the proposed method can both remove the noise completely and reserve image signal extremely, and it is applicable to filter images contaminated by different densities of salt and pepper noise.(3)Based on the characters of singular value of image matrix, this thesis proposes a new contrast enhancement method of tire and tire trace images which combines frequency division with singular value enhancement of matrix. The method divides the image frequency via Butterworth low pass filter, enhances the low frequency part by means of singular value enhancement method, and enhances the high frequency part by the linear enhancement method. Experimental results demonstrate that this method performs well in enhancing detail and information entropy of low contrast image, high contrast image and fuzzy image.(4)In this thesis, the texture character of tread pattern image is extracted by a new feature extraction method integrating gray level co-occurrence matrix(GLCM) into non-subsampled Contourlet transform(NSCT). The effective features are selected according to the threshold interval between image features, and the optimal decision-tree-based support vector machine(SVM) classifier is constructed in accordance with the separable degree of these effective features in order to achieve the classifying identification of tire pattern images. Experimental results demonstrate that the features extracted by the feature extraction method and feature selection method have higher separable degree, and the classifying identification method based on the optimal decision-tree-based SVM can achieve a high recognition rate.