Research On Vehicle License Plate Recognition System Under Complicated Background

With rapid development of road transportation in China, the System of Intelligent Traffic Management (ITMS) has become the focus of attention. Vehicle license plate recognition system which is a part of the ITMS has been widely used in many areas, such as the bridge crossing automatic fare collection, recording the crossroads' illegal vehicle and so on. The recognition technology has been studied for a long time, however, there are still some problems in existing algorithms. For example, the environmental adaptability is poor and recognition accuracy rate is relatively low in complex background. Therefore, the research of license plate recognition algorithm is still a hot topic and has practical meaning.A comprehensive study in main technical parts of vehicle license plate recognition system has been done in this paper. Based on the research at home and abroad, we completed the following three main areas:Firstly,a license plate image pre-processing method has been proposed, including illumination compensation, graying, wavelet denoising, Sobel edge detection operator, binarization. Based on this methods, the plate has been exacted from the car using geometric and texture features. According to practical requirements, the method of edge detection has been improved in this article, which is divided into horizontal and vertical directions. The results show that the method positions accurately and has small unrelated region.Secondly, there is a problem that one character is divided into multiple characters using simple vertical projection. To solve this problem, the article use expansion and corrosion to highlight the plate feature, and add a criterion to distinguish which one needs to be merge or which one needs to be re-partition. This method allows a more accurate segmentation.Lastly, recognition rate is relatively low for the single template matching and neural network. In this paper, neural network and template matching are combined effectively. Further more, the new idea of characters' central pixel weighted using thinning algorithm on the basis of the original template matching is proposed. The experimental results show that this method proposed here is more suitable for practical applications.