Study On Evolution Of License Plate Characters SIFT Features Features In Image Degradation

SIFT algorithm is one of the hot and difficult research areas of image featurematching at home and abroad in recent years. The thesis is focused on the evolution oflicense plate character SIFT features in image degradation based on the introduction ofthe method of SIFT features extraction.The license plate characters contain rich SIFT features. Therefore stable SIFTfeatures can be extracted from license plates with different sizes, angles and scales. Firstsimulated data is used for experiment. A few characters are selected from standardlicense plate character library to form into a simulated license plate. Anthropogenicdegradation is added to this simulated license plate. Here the impact of adding whiteGaussian noise, down sampling and border on image is mainly studied. Features of thestandard characters and the processed images are matched. Record the results, fromwhat we can get the evolution of SIFT features in image degradation. Therefore thetolerance can be obtained, that is, to what degree the image is degenerated when we stillcan ensure the existence of image SIFT features. Experiments show that for simulatedlicense plate sized207*68adding the25.99%white Gaussian noise，that is with themean0and the standard deviation130, and reducing its scale twice as its tolerance.Then, the experiment is repeated using various real license plate images for furtherconclusion. License plate image is intercepted from the video captured by cameras,normalized to the size207*68and then processed by adding white Gaussian noise anddown sampling. And its SIFT features are obtained and matched with those of standardcharacters in the standard license plate character library for further conclusion. That is,for real license plate image sized207*68, adding the12.46%white Gaussian noise，thatis with the mean0and the standard deviation90, and reducing its scale twice is itstolerance. Finally the thesis introduces the application of SIFT algorithm in license plateimage restoration. First license plate images are collected by photographs or video andafter preprocessed for fast license plate location to get the license plate area. Then thelicense plate characters are divided and identified. Finally output the clear license plate.Experiments show that the method using the SIFT feature matching of license platecharacters is effective for image restoration.