A dynamic template combination of pixels for license plate recognition

Abstract

This invention creatively puts forward a method that we can use a dynamic template combination of pixels for character segmentation in license plates. Firstly, we can obtain the dynamic vehicle monitoring area through the background modeling based on Gaussian mixture model and local license plate by using the edge information of license plate. Secondly, combining structure information, texture information, the rules of arrangement and maximum correlation among characters, we will find segmentation position after sliding this template. Then we can get the best position of license plate segmentation through calculating the number of nonzero pixels in the segmentation position in a binary image. At the end, after extracting the features of the characters, we should use the SVM for training and license plate character recognition [1, 2]. Compared with other solutions in the field, the invention has high accuracy, good robustness and stability in complex environment.