Research And Imple Mentation Of License Plate Recognition Algorithm Under Complex Background

With the rapid development of economy and technology,the pressure of urban traffic management is increasing constantly.Intelligent transportation has become a strong demand with social progress.So the license plate recognition as one of the core became the important research.The license plate recognition mainly includes three modules: license plate location,character segmentation and character recognition.Although the current research has reached the level of availability,but most of them can achieve better results in the fixed conditions only.On the other hand,with the continuous increasing of the number of vehicles,the application of license plate recognition scene is increasing.And more and more problems is found.Fog can easily lead to blurred license plate image and influence license plate location.Plate stain and fading can easily lead to character adhesion and influence character segmentation.Unfixed Shooting angle can easily lead to character inclining and influence character recognition.Firstly,this paper proposes a license plate location algorithm based on character edge point extraction.This algorithm combined with color and te xture features,and add edge extraction,so it can get a more accurate result.The goal of plate location is to find the area of the license plate from the image containing and cut it out.The algorithm is affected by the size of the image,the illuminatio n condition and the image definition.License plate has a certain texture characteristics,such as it has seven characters,their size is same and the space between them are fixed in our country.The color collocation between the license plate background and the character is fixed.So this paper used a method combining the texture feature and the color feature and get a better result.Secondly,character segmentation.In order to get the independent character image,it must overcome the difficulty caused by the stain interference,the character adhesion and the character tilt.In this step,taking into account the problems caused by the tilt will lead to the difficulty of the subsequent steps,this paper made horizontal and vertical correction to the image using the method of rotating projection,and then remove the license plate frame.In order to solve the problem of split failure caused by the stain influence and character adhesion,this paper used a method combining template matching to vertical projection.And the method change the width of the character dynamically,so that the width of the character can be estimated more accurate.At the same time,the character boundary in the horizontal direction is divided Independently,so the result is more accurate.Thirdly,character recognition.The focus of character recognition is recognition rate and time performance including two types of methods based on SVM and neural network.This paper analyzes and illustrates the advantages and disadvantages of the two kinds of methods,and makes pretreatment and feature extraction to the image.Then using the improved BP neural network,it designs three kinds of classifier including Chinese,letter and number.And it compares the performance of different neuron nodes.In the experiment,it compares the recognition method based on SVM and RBF neural network.The results show that the recognition method based on BP network has a high recognition rate,and the performance is more stable under the interference of the stain.