Research On Chinese Character Recognition Of Low Quality License Plate

License Plate Recognition(LPR)system, a key link to achieve Intelligence Transportation System, is based on a series of technology like computer vision, image processing and pattern recognition, etc., to extract vehicle license information from image. The study of Vehicle License-Plate Recognition technology has great and practical significance and wide application prospects.Character recognition is a core component of Vehicle License-Plate Recognition system. According to the current License plate style system of our country, Character recognition mainly includes three aspects, English letters character recognition, Arabic numerals character recognition and Chinese character recognition.This paper, through the consult massive literature material, has studied Character recognition technology, emphatically settled a series of difficult problems in Chinese character recognition and proposed two related algorithms.This paper has proposed Chinese character recognition algorithms based on fractal dimension and 2d discrete wavelet transform. First, to pre-processing vehicle license image; Second, according to the given length-width ratio of vehicle license to extract every part of Chinese character as needed, and to form square images which are necessary to fractal dimension algorithms, then to successively extract the whole character image, local image and fractal dimension of components in each direction of wavelet transform, finally to form them into feature vector; third, to put feature vector into Support Vector Machine(SVM) for classification recognition. The experimental results demonstrate that due to the high self-adaptability of fractal image, fractal dimension identification method can successfully recognize vehicle character image with stains, color fading, conglutination and fracture. With the combination of 2d discrete wavelet transform, better interference suppressors can be achieved. The character feature vector based on this can improve the recognition rate more efficiently.The paper has also proposed a license plate character recognition algorithm with the combination of Wavelet Transform and Kenel Principal component analysis (KPCA). The methods of this algorithm are still to launch feature extraction, then to identify the characters with SVM grander. However, in the section of feature extraction, this algorithm first to operate the character images in sub-block, then to stikeing to two level Wavelet coefficients of every Sub-image. Finally, with the matrix integration method, fusion the block characteristics to the required eigenvectors.The experimental results demonstrate that, compare with the former algorithm, this algorithm is more efficient and more robustness to the blurred images.