Research On License Plate Location And Character Segmentation In License Plate Recognition System

As an important research topic of intelligent transportation system,license plate recognition system mainly includes three key modules: license plate location,character segmentation and character recognition.Among them,the license plate location and character segmentation directly determines the accuracy of the character recognition.Based on the research and analysis of domestic and foreign scholars in the license plate location and character segmentation part of the latest research results,this paper mainly studies the license plate location and character segmentation algorithm.The main work and research results are as follows:1.In the license plate location algorithm,this paper puts forward a license plate locating algorithm based on a simplified Pulse Coupled Neural Network(PCNN)and comprehensive features combined with Chinese own characteristics of license plate.The algorithm consists of three steps: image preprocessing,locate license plate roughly and precise positioning of the plate area.For the color image data quantity is large and the light illumination is insufficient when the image with low contrast problems,we do gray,top-hat transform and gray stretch preprocessing to the collected license plate images.Because the license plate has the characteristics of dense edge,we use Sobel operator for edge detection on the image,achieve the purpose of highlighting the license plate area.During the process of the coarse position of license plate,we use mathematical morphology method to get the license plate candidate region,and then use the structure,color and texture features of license plate to filter the candidate region,overcome the shortage of single feature location algorithm.In front of the accurate positioning,in view of the high speed motion blurred images,we use a simplified PCNN for license plate binarization and do tilt detection and correction of the license plate.Finally using texture jump characteristics of license plate area,we adopt the method of line scan and vertical projection combined with color features to precisely locate license plate.2.In the license plate character segmentation algorithm,firstly preprocessing of the located license plate,including the normalized and gray enhancement of license plate image,we present a normalization method based on license plate color.In order to further improve the character segmentation accuracy and overcome the shortage of the existing algorithm,this paper proposes a license plate character segmentationalgorithm based on connected domain,vertical projection,recognition feedback and prior knowledge.Because the license plate separator area has unique characteristics different from other areas,we should locate the separator area,and then reference the separator area and combing with the characteristics of the license plate characters to segment the last five characters of separator area by using the character segmentation method based on the combination of connected domain,vertical projection and recognition feedback.After getting the standard character width,the paper uses the method of combine the vertical projection with prior knowledge to segment the first two characters.Finally,the thesis meets the needs of the recognition by doing normalization processing.The algorithm of this paper do experiment under Matlab2013 programming environment,and other algorithms of the literature in the same experimental platform and data set for comparative analysis.Experimental results show that the proposed positioning and segmentation algorithm under different environment can get higher accuracy and can meet the needs of practical application.