Chinese License Plate Localization in Multi-Lane with Complex Background Based on Concomitant Colors

摘要

Color is an important feature of the Chinese license plate. In this paper, a license plate location method totally based on color information for the most common blue and yellow plates is proposed. First, use a special way to generate a grayscale image, highlighting the blue and yellow regions of the original image. Then by using the color concomitant property and transitions between the background and the characters, fake plates are removed, and real plates reserved. Finally, use the color information again to get a more accurate location result and remove the fake plates by using the relationship between the stroke width and the character size. The proposed method adopts two kinds of threshold methods: global threshold method and adaptive block threshold method. By automatically judging the illumination distribution of an image, the algorithm can decide which to use. The proposed method can be used to locate license plates in multi-lane with complex background, and works well even on images with uneven illumination. 2207 images from different traffic videos are used to do experiments, including 1314 images of daytime and 893 images of nighttime. Each image contains 0-7 plates, and the location rate achieves 95.01%.