A Fast Method for Two-stage Precise License Plate Location Based on Fusion of Corner Feature and Gray Level Jumps

Because the characters on plate license have enough corners which are distributed intensively, in this paper, a method of plate location based on fusion of corner feature and gray level jumps is proposed for license plate location which has higher accuracy and efficiency than most existing popular methods in literature. The method is mainly divided into two stages involving rough location and accurate location. On the first stage, an improved Harris corner detection is used to detect the corners with much less time. Then, only one candidate region is chosen according to the intensity of corners and the color of plates. On the second stage, the license plate is located accurately using K-L expansion based plate tilt correction and boundary scanning on account of gray level jumps. Experiments on 201 real license plate images have been conducted, 198 plates are correctly located and the average time consuming is 59ms which is much less than most existing popular methods.