Research On License Plate Segmentation Under Complex Background

With the developing of information technology and intelligence technology, theinformatization and intelligentizing of traffic management is the trend. License PlateRecognition system (LPR) is the core of Intelligent Traffic System (ITS). It is veryimportant in modern traffic management systems.In recent years, use the computer to deal with and analyses image has already obtainedprogress at full speed. The study on the critical techniques for LPR has already become animportant research field of scientific circles.In this paper we discussed the three parts of License Plate Recognition Systems: imagepre-process, license plate location and character segmentation, and made a deep researchabout some important and key technology.A hybrid license plate segmentation scheme is presented in this paper. The mainlythree stages of this approach are designed to deal with images taken under various realworld conditions. Generally, the images of complex background, license plate skew anddifferent plate size are used in our experiment.At first, in the license plate roughly location stage, we proposed a candidate checkingmethod based on texture features of license plate because the texture of plate area is muchricher than non-plate areas. In the license plate extraction stage, BP neural network are usedto verify the real license plates, after multiply features of the license plate are extracted baseon the two method of PCA and texture feature analysis. In the last stage, we use Radontransform to do skew correction, and then remove the noises and segment the characters.Based on the experimental results, it proves that this proposed method can relativelylocate license plate and segment characters and the performance of the system is promising.It shows that combined feature extractions and segmentation techniques can improve theability of system, and the combination of neural network and artificial intelligence under theprinciple of efficiency and practicality will be major development tendencies of patternrecognition.