License Plate Character Segmentation Research And Implementation For Low-quality Multi-scale Image

With the rapid development of national economy, the quantity of our country automobile is rapidly increasing, while the highway traffic is facing more and more heavy burden; since intelligent transportation system (Intelligent Transportation System, ITS) can make scientific, efficient use of existing transportation infrastructure resources, it is becoming the main method to further mining traffic of our country city ability. License plate recognition technology is one of the domain application and research topics of the computer pattern recognition technology in intelligent traffic. Typical license plate recognition system mainly includes the following three steps:license plate locating, character segmentation and character recognition.This paper studies the vehicle license plate character segmentation for multi-scale the low quality imgaes. Firstly, this paper briefly expounds the present situation of license plate character segmentation research. In view of the current algorithm cannot accurately segment the adhesion in license plate character areas and cannot accurately selecting out the license plate characters from shortcomings the candidate region, this paper puts forward a multi scale segmentation algorithm based on prior knowledge. The experiments show that the algorithm is able to segment license plate character on the rough positioning plate.On the study of binarization, the image processing by global thresholding method is easier to segment, but it can’t handle the uneven illumination of the license plate; local threshold method can deal with the uneven illumination of the license plate, but the normal light shooting license plate after processing by this method will appear more noise area which make it much harder to segment. This article considering the advantages and disadvantages of global thresholding and local threshold method proposes an adaptive algorithms of binarization.On the study of the license plate segmentation method, we take the strategy of basing on the connected component of the segmentation. In the adhesion character segmentation process, this paper proposes the algorithm of water drop to improve the shortage of the existing segmentation algorithm. For the selecting of license character region, according to China’s license plate feature, we put forward a kind of new priori embedding method based on Bayesian belief network.. For a given vehicle, the probability coding of each candidate region of the license plate character can construct Bayesian belief network, which will convert the problem from the selecting of license plate region into the construction of Bayesian belief network topology.The experiments show that the method can accurately not only screen communicated Chinese characters, but also remove the interferences such the left and right frame, as well as the light of the car.Finally, this paper developes a multi scale low quality license plate character segmentation system, and gives the related experimental results and data.