Research Of License Plate Recognition System Based On Digital Image Processing

This paper has studied in detail the characteristics of Chinese license platerecognition system, divided into three parts: the license plate image location, licenseplate character segmentation, license plate character recognition.The license plate location: algorithm based on gray transitions andmorphological characteristics of the adaptive binarization method for automotiveimage binarization. Since the license plate in the horizontal direction of the graychange, significantly higher than the background, so you can preprocess the imageaccordingly; reuse morphological opening and closing operation, eliminate imagenoise and fill the image area voids, to provide help for the license plate location.Based on this, we use an adaptive binarization method: first order differential imagein each column is multiplied by the average of the maximum value of0.5as theinitial threshold Th, the image binarization; Then the image morphologicalprocessing; Check the binary point of the total white area in the figure on thediagram, if less than the given value decrease the value Th, on the contrary increasedif repeatedly revised threshold for image binarization again, and the operation foropening and closing operation in order to obtain the best results. After such aprocessed image on the plate area is very obvious; then extracting candidate regionsthrough regional analysis, combined with prior knowledge remove the pseudo-areamethod to achieve the license plate fast and accurate positioning.Character segmentation: The plate precise cut. Since the projection of thehorizontal plate having the maximum gradient value of two projection view,corresponding to the character frame and the gap between the upper and lower, it canbe removed which set the upper and lower border of the upper and lower boundariesof the character; then the vertical projection of the plate, projection peak at thebeginning of the leftmost and rightmost end of the peaks corresponding to thecharacter of the left and right boundaries, thus completing the accurate segmentationof the license plate, and on this basis, the character segmentation. Segmented plateimage for accurate vertical projection valley between two peaks is split position. Inorder to avoid a segmentation fault, the extracted "character" in checking the numberof pixels, the total experience determined character area which is normal, noise, or anarrow area, and depending on the narrow area adjacent regions checks to determinenarrow characters or misuse segmentation, thereby performing a narrow regionsegmentation process to obtain the correct result.Character Recognition: In order to avoid noise, this non-linear normalizationmethod adopted, the character is normalized to the size of the character pattern of thesame32×16pixels. Then choose a template matching method for the license plate character recognition, a character is about to split, respectively, with the charactertemplate characters differencing operation, with the smallest difference about whichcharacter image segmentation is determined that the characters and theircorresponding character output to complete the identification.In this paper, Matlab as a tool to carry out experiments on a variety of differentangles, different environments license plate image, the experiment results show thatthe algorithm is designed in accordance with this license plate recognition system toimprove the accuracy of license plate location, the recognition rate has increased, butalso the processing speed has been improved to some extent.