Application of intelligent systems for Iranian License Plate Recognition

Despite recent advances in Vehicle License Plate Recognition Systems (VLPRS), there are still several challenges in these systems such as incompatibility with different conditions. These incompatibilities are caused by some reasons including shadows, skews and dirt on license plate, changing illumination intensity, weather conditions, varying distance between camera and vehicle and rotation of license plates. This paper investigates the challenges related to shadows on license plate, changing illumination intensity and other similar cases. They will be solved by the Bernsen thresholding method along with other new techniques. Moreover, the proposed approach tries to provide a rotation and size invariant system. It consists of three stages: plate Localization by horizontal projection and Sobel filter, character segmentation stage through vertical projection, and character recognition by 4-directional distance profile features. This paper uses a database containing 400 images of vehicles under complicated and non-uniform conditions whereas 200 images for training and 200 images for system evaluation. The accuracy rate obtained in the above three stages is 88, 85.6 and 96.02 percent respectively.