An efficient algorithm for Chinese license plate location

This paper presents an efficient and robust method of Chinese license plate location. In this method, firstly, 2-D wavelet transform is implemented to a digital car image to detect the high frequency information which is the basis to the next approaches. Secondly, mathematic morphology is used to get connected regions which contain the candidate plate regions with appropriate size. In order to pick out the real plate region, a BP neural network is trained by the samples chosen from plate images under different illumination conditions. Then certain of pixels in candidate regions are identified by the network based of color elements, which help determining the real plate location. Experimental results demonstrate the efficiency of our algorithm.