A Study On The Character Recognition Algorithm Based On Neural Network

Artificial neural network is an important research field of machine learning. It is a calculation model which based on the simplification, abstraction and simulation of brain's neural network. The artificial neural network ensemble constituted by multiple sub-networks can significantly improve the generalization ability of weak learning model, considered as an engineering neural computing method with a broad prospect.The artificial neural networks can be applied to character recognition, face recognition and other fields. In this paper, LPR (License Plate Recognition, LPR) for the study, we explore the character recognition based on neural network algorithm application. LPR is an important research content of the Intelligent Transportation System (ITS), composed of license plate location, character segmentation, feature extraction and character recognition. License plate location and feature extraction is a prerequisite for character recognition. This study focused mainly concentrated in the license plate location and the efficiency improvement of ensemble classifier. The difficulties of License Plate Recognition are lower vehicle license image quality, characters distortion, and small proportion of license plate image, mass noisy in license and so on. In response to these difficulties, this paper locate the license plate position based on variable window scan, the wavelet transform, edge detection and line scanning method. It can overcome the light, color, size and position of influence and efficiently detect the plate. For the tilt of the license plate, this paper correct the plate stably and fast, using Radon transform based on image projection. The license plate character segmentation uses vertical projection method combined with the prior knowledge, quickly and simply. In the character recognition module, respectively, the improved BP network and ensemble classifier are used. Finally, dynamic adaptive weight cutting methods is proposed to improve the AdaBoost efficiency.In this paper, the DAWT AdaBoost can improve the time performance of the AdaBoost ensemble classifier while works with a large number of training data. At the same time, DAWT AdaBoost compares with the traditional AdaBoost, the static weight trimming AdaBoost and dynamic weight trimming AdaBoost. Experimental results show that the DAWT AdaBoost improves the class precision and time performance, worthy of further study.