Research And Implement On License Plate Character Recognition

Intelligent transportation system (ITS) plays an increasingly important role in moderntraffic management. As the crucial component of intelligent transportation system, licenseplate recognition (LPR) is a very important research issue in intelligent transportationsystem. License plate recognition system uses the knowledge of computer vision, imageprocessing and pattern recognition to recognition the license plate, it has a very largeapplication prospect.License plate recognition is divided into three major phases: license plate location,character segmentation, character recognition. License plate location as the first stage oflicense plate recognition, locate the license plate mainly based on the edge information ofplate area. Because there are abundant edges in plate area, we use improved sobelalgorithm to extract edge, and then edge density analysis method to locate the plate regionaccurately and quickly. Character segmentation is to get the character image for characterrecognition after the plate location. Because the characters are ranged regularly and thedistances between each characters are all most the same, first we binary the license plateregion, then use the connected component analysis method to search the markedconnected component, we can determine the size and position of the characters based onthe position and size of the connected components.Character recognition as the final stage of license plate recognition system, therecognition result directly determines the quality of license plate recognition system, thecharacter recognition algorithm needs to deal with the impact of inaccuracy that locate theplate area and segment character, the algorithm also needs to control the false detectionrate of the whole system. For the ordinary character recognition, multi-template matchingalgorithm is used to recognize character quickly and accurately, neural networkrecognition algorithm is used to recognize the similar characters and Chinese characters.Finally, we treat the cross-correlation operator of multiple template matching algorithm asthe confidence degree of recognition, calculate the average confidence degree ofrecognition, remove the pseudo license plate based on the average confidence degree ofrecognition, obtained desired results. License plate recognition system has a very high real-time requirement, the method inthis paper take the recognition accuracy and real-time requirement into account. Theexperimental results based on1000images shows that the algorithm that proposed in thispaper can locate the plate region quickly, segment the character image accurately, and thealgorithm has a fast compute speed. This license plate recognition has a bright applicationprospect.