License Plate Recognition Algorithm In The Complex Context Of Research

In recent years, with a significant increase in the domestic motor vehicles, the research and development of license plate recognition systems become increasingly important. For the license plate recognition technology, there are many new algorithms and design ideas, but there are still many technical difficulties. For example, restoration of images which are taken motion blur due to the high speed, license plate location and segmentation in complex background, tilt correction plate, characters recognition which is polluted and easily confused. This topic proposes some practical algorithms for the existing difficulties.This review focuses on the motion blurred license plate restoration, license plate localization, character segmentation and character recognition. The specific Algorithms are described as follows:1) Motion blurred license plate restoration. Motion blur make the image information extraction difficult, its recovery has important practical significance. An improved motion blur restoration algorithm is presented in this paper. The improved high-pass filter fast algorithm is used to determine the direction of motion blur. Edge detection algorithm is used to determine fuzzy scale. Finally, the blurred image is recovered by Wiener filter. A new algorithm is proposed for automatic estimate K value in this paper.2) Vehicle license area position. A new algorithm of vehicle license plate location based on morphology-line scan is proposed. This algorithm has good performance for vehicle license plate location in low-quality images. Firstly, the plate edges are determined. Secondly, morphological operations are done to get several connected region, and then license plate is rough located based on plate aspect ratio. Finally, line-scanning method is used for precise positioning according to the number of black and white pixel transform. Then tilt plates are corrected by the Hough transform.3) Character segmentation. First of all, the frame in license plate is removed and the image is normalized. It makes characters segmentation very simple. With license plate features, the dimension of characters and character spacing in normalized image are calculated. The segmentation points of each character can be determined easily by the above step. The effects of the environmental factors are removed by this algorithm. The success rate of this algorithm is high.4) Character recognition. Using BP neural network, segmented characters are directly stented into the network. Features of Character are extracted independent by network. Trained network parameters are saved and are read when the character is recognized. The results of recognition are output.