License Plate Recognition Based On Derived Kernel Model

License plate recognition systems are now used in traffic control system, parking fees, vehicle tracking and many other fields. This system mainly refers to license plate location, character segmentation and character recognition.For license plate location, this thesis first transforming shot image to grayscale, then utilize Sobel operator to extract the edge information in order to intensify the image contrast for stretching the gray level. Then the edge information could be processed through using corrosion in morphology and erosion operator, hence license plate region could integrate, and eliminate interference of micro-region. Finally, the vehicle license plate is located by its grounding-color recognition.For license plate character segmentation, the gray image of license should be transformed into binary image. In order to remove image noise, the morphological operator could make the characters more clear. In domestic, the standard license plate is composed of seven characters without any adhesion. The left and right borders are determined based on the vertical projection of the characteristic image. As the vertical area is less at the boundary of character, some of which reach to 0, line scanning method should be used through setting a threshold value to split out of seven characters, and then normalize the characters.For character recognition, the method of exporting of nuclear model recognition could be utilized. Based on the three layers of nested structure, the template selection standard criteria could go forward. Using slide match to calculate neutral responses and derived kernel. Based on the maximization of derived kernel,the characters could be recognizes. Comparing to the method of Support Vector Machines, derived kernel has the characteristics of scale-invariance, rotation-invariance, translation-invariance and higher computing speed. Through analyzing influencing factor of the recognition rate of derived kernel, it could be adjusted by changing the size of first and second layer templates.