Research On Segmentation And Recognition Of Fuzzy License Plate

License plate recognition technology is an important research topic in vehicle control research.At present,it is widely used in highway toll collection,parking management,traffic road,vehicle scheduling,vehicle detection and so on.Although the license plate recognition technology has been mature at the present,the recognition accuracy of fuzzy license plate is not ideal under the influence of complicated conditions such as blurred image and light change.This topic is one of the key technologies of a new product research and development in the an enterprise commissioned project,and we will focus on the problem of positioning,segmentation and identification of fuzzy license plate.The main work of this thesis are:1.Location of license plate image:A method of vehicle image location based on image color,edge feature and MSER text search method is proposed.Through the digital image processing technology and the depth of learning network model to bayonet license plate image as the input data,automatically learn the license plate characteristics and obtain license plate position.2.Fuzzy License Plate Segmentation:In this paper,we propose a method of image segmentation based on contour method and projection method,which includes the secondary location of license plate,background foreground segmentation,license plate threshold processing,license plate projection segmentation and so on.3.License plate character point extraction and classifier selection:This paper presents a combination of projection histogram features + artificial neural network and HOG feature + support vector machine license plate character image,named double classification method.The two characteristics of the binary image of the license plate character are extracted respectively,and the classification is predicted by using the double classifier.4.Special character recognition method:In this paper,we propose a local feature region matching method for fuzzy,clustering and missing of license plate character image,and make a second judgment on the special characters of the double classifier classification.Based on the above research,this thesis constructs a complete recognition system of license plate characters.The training data set used in the experiment comes from more than 6000 license plate character images collected by enterprises on the actual road,including 31 Chinese characters and 34 English alphanumeric characters.We have identified more than 13,000 actual license plate test data sets,the results show that the license plate character recognition accuracy rate of 95.1%.