Research On Segmentation Of Degraded License Plate

With the development of information technology and artificial intelligence technology, the informatization and intelligentization of traffic management system is a general trend. The license plate recognition (LPR) system is the core of the intelligent transportation system (ITS). It is very important in modern traffic management systems.The LPR system consists of three steps and five key techniques. The steps are license plate location, character segmentation and character recognition. The key techniques are image preprocessing, license plate location, tilt correction, character segmentation and character recognition. The methods of license plate tilt correction and character segmentation are studied in this paper.In the process of obtaining the license plate images, due to the angle change between the camera and the plate, the vehicle image usually tilts. This phenomenon leads to the distortion of the plate and the characters, and makes it difficult to the character segmentation and character recognition. A novel approach for vehicle license plate tilt correction based on Radon transform is proposed. Firstly Radon transformations are performed on the license plate image, and then the sum of absolute difference of the results is calculated, lastly the tilt angle is obtained by located the maximum. Experimental results show that this method is robust to the dirty license plates and the license plates under various lighting conditions, and it is practical.As a key step of the LPR system, Character segmentation is the foundation of character recognition. The results of character segmentation determine the accuracy of recognition directly. This paper proposes a novel method for character segmentation of degraded license plate based on prior knowledge. Firstly character segments are located according to the vertical projection, and then the character segments are merged or split according to the vertical projection and the prior knowledge. This method is more efficient under the condition that the license plate images are degraded, such as uneven illumination, poor contrast, tilt, discoloration, fading, broken character, connected character, and etc.Based on the standard license plate segmentation, this paper analyzes the characteristic of double rows license plate and proposes a method for double rows license plate segmentation based on Gaussian fitting. Segment position between the double rows is located by using Gaussian fitting, and then the characters of the double rows are segmented by using the standard license plate segmentation method. Experimental results show that this method is effective to segment the double rows.