Research On BoW In Image Recognition System And Defect License Plate Recognition Based On Real Rough Set

License plate recognition has a widespread range of application prospects,which is one of the hotspots in the field of pattern recognition.The traditional pattern recognition method has certain limitations on the license plate recognition because of the similar shape,font fragmentation,font fouling and license plate damage in some vehicle license character images.This paper constructs a license plate recognition decision information system based on the image recognition model,the real rough set theory and the variable granularity principle.Through the image recognition system,the real rough set theory is used to analyze the image features.This paper explores an effective way to improve the accuracy of defective license plate recognition by variable granularity simulated feedback mechanism.The main work of this paper is as follows:1.A method of building license plate decision information system based on real field set is explored.The characteristics of license plate features are merged into a set of high-dimensional eigenvectors as conditional attributes based on the real rough set theory,and the license plate recognition decision information system is constructed to solve the problem of traditional roughness.The problem of inconsistency caused by the interference of a small amount of noise data in the set method affects the whole classification relation.2.A method of attribute reduction of license plate based on real rough set is presented.According to the real rough set theory,the attribute importance of the license plate characteristic attribute is defined,and the classification ability of the license plate characteristics is defined by the attribute importance of the license plate characteristic.The license plate feature attribute reduction algorithm based on real rough set,which makes up the shortcomings of the traditional rough set for the importance of the attribute is not clear,can not effectively reduce the shortcomings of the appropriate characteristic attribute according to the importance degree of the attribute,and effectively reduce the license plate characteristics in the license plate recognition decision information system.3.A mechanism for determining the variable size of the license plate is established.According to the theory of particle size,the granularity of cognitive information needed in the process of license plate feature reduction is defined.The training sample feature set is improved in the previous feedback,and the optimal size of cognitive information is obtained.It is difficult to achieve human cognitive things from coarse to fine,from the simple to the level of repeated collection and processing of information process by the traditional cognitive system unidirectional open-loop way.The vehicle license intelligent cognitive algorithm presented based on variable granularity simulated feedback mechanism solve this problem,so as to improve the correct rate of license plate recognition.4.A rule fusion method for license plate characters is proposed.A two-layer classifier is designed to fuse the decision rules of the license plate recognition decision information system after the reduction of the decision rules which can not be uniquely matched in the process of license plate recognition.Thereby extracting the true attribute of the bag.According to the theory of recursive convolution neural network,the sufficient condition of the limited information capacity is established,which solves the problem that the defective license plate image itself is similar to the shape,font fragmentation,font fouling and license plate damage.In order to verify the advantages of the fuzzy recognition algorithm based on variable granularity feedback mechanism and the rule fusion mechanism of license plate characters,this paper adopts 1402 categories of 45 license plates in the license plate image database published by Sun Yat-sen University OpenITS platform license plate image samples to Matlab,Visual Studio 2013 software as an experimental platform,the algorithm proposed in this paper was verified.The experimental results show that the proposed method is effective and feasible.