License Plate Classification from a Binarization Perspective

For binarized license plates, there are usually two categories: one is the white-character type, of which the character is white and the background is black. The other is the black-character type. For some linearization-based algorithms, such as character segmentation, we should apply suitable approach according to different linearization results. It is still a challenging task for multi-style license plates. In this paper we propose a stroke-width-transform-based method for this problem. Firstly, we calculate the stroke width transform by using the original grey image and the inverted one, respectively. Secondly, the histograms of the corresponding stroke width transform images are generated. Thirdly, the image, corresponding to the maximum value of the histograms, is selected. Finally, if the original image is selected, the license plate is the white-character type and vice versa. The experimental results show the proposed method is superior to others by using multi-style license plates in the United States, but the time complexity is high.