Study On Various Color License Plate Recognition System Based On Artificial Neural Network

With the fast development of the domestic economy,the social information level has increasingly improved,and the intelligence of the transportation system has become a developing trend.License plate recognition system is a important component of the intelligent traffic management system.It can be used in many vehicle management places to save manpower and material resources.Therefore,it has extensive application prospect.The automatic vehicle license plate location and recognition system is a important and fundamental problem in this area.The license plate location and recognition system mentioned in this paper consists of four sections:image preprocessing,license plate location,character segmentation and recognition.There are two main tasks for the image preprocessing:first,adjusting the luminance curve dynamically in order to cope with the luminance change caused by light;second,using gray image top-hat transform to eliminate background noises according to the bring-dark alternation on license plate area.In location process,in order to make the system can work in complex backgrounds and have real-time ability simultaneously,the system is designed to use basic mathematical morphology to location at first,and then use complex texture feature location method if the background is too complicated.After that,license plate color judgment and skew correction are needed.In character segmentation,compared the traditional vertical projection and template matching method,a modified vertical projection method combined with prior knowledge is presented here.It can not only correct the location result,but also identify "1" before recognition process.In character recognition,SVM and BP networks are combined together and achieve a better effect than using them respectively.The difficulties in developing focus on location and recognition processes.It is hard to locate the license plate area since the complex backgrounds and various colors on license plate.The location error will cause problems for segmentation and recognition behind.The recognition problem is mainly caused by the irregularity of the characters.The system was simulated on Matlab and then translated by VC6.0.These two versions can help for each other,find out problems and search better algorithms.Finally,the system can work smoothly and achieve the expected assumption.