Research On Super-resolution Restoration Of Color Vehicle Plate Image Sequences

The aim of super-resolution restoration technology is reconstucting ahigh-resolution non-aliased image(or high-resolution non-aliased image sequences)from low-resolution aliased and deformable image sequences for removing additivenoise and blurring produced by the finite detector size and optic element by signalprocessing method, and suppling the low-resolution limitation for practical hardwareand finite cost, and effectively improving the quality of degradation image in theprocess of imaging.The method can improve image resolution,which is characteristicof economy and easy realization. Intelligent traffic management has become the maindevelopment direction of traffic management, and vehicle license plate recognitiontechnology is the key technology of intelligent transportation system. Sosuper-resolution restoration of image sequences of color vehicle license plate is worthresearching, which is related with vehicle license plate recognition. The primaryresults of this paper are summarized as follows：Blur identification methods and deblurring methods of vehicle license plateimages are researched in the paper. Traditional deblurring method and the theoryknowledge of support vector machine are introduced. Different parameteroptimization methods of SVM are comparatively analyzed. Blur parameters of blurredvehicle license plate images are identified by SVM, then blur types of blurred imagesare classified. Different restoration methods are used to deblur vehicle license plateimages. The subjective and objective evaluation standards are used to evaluaterestoration images by different restoration methods.Motion estimation between both image sequences is researched. The principle ofimage registration, registration model and many block-matching algorithms are introduced. Block motion estimation mainly includes the selection of the block shapeand size, block matching criterion and block matching algorithm. Considering theparticularity of the color space, different block-matching algorithms are used toestimate the motion between vehicle license plate image sequences in this paper,which use the matching criteria of mean square error and the matching criteria ofaverage absolute error in gray space, the RGB color space and HSI color spacerespectively. Based on experimental data, the performances of block-matchingalgorithms are comparatively analyzed in different spaces and in the condition ofdifferent block matching criteria in order to obtain that the accuracy of motionestimation of color images is related with color space and search criteria. The bestmotion estimation method of color vehicle license plate image is also obtained.The super-resolution reconstruction algorithms of image sequences are tested inthe paper. Super-resolution reconstruction model and super-resolution reconstructionalgorithms of image sequences are introduced. Four different reconstruction methodsare used to reconstruct a super-resolution image from deblurred image sequences.Super-resolution reconstructions from image sequences are comparatively analyzed.Information entropy, average gradient and reconstruction algorithm time are used toanalyze the performances of four super-resolution reconstruction algorithms of imagesequences. A super-resolution image from blurred image sequences is alsoreconstructed. The differences between the super-resolution images from blurred anddeblurred image sequences are analyzed to obtain the necessity to deblur imagesbefore reconstruction.