【作者】 沈世旻；

【导师】 盛翊智；

【作者基本信息】 华中科技大学 ， 控制理论与控制工程， 2004， 硕士

【摘要】 车牌识别技术已经被越来越广泛地应用到现代交通的各个领域,在提高工作效率、降低劳动强度方面发挥着积极的作用。本文以“车牌自动检测系统”项目为背景,研究了计算机视觉和模式识别的相关理论,设计出车牌识别的算法,并应用到生产实践当中,最终实现了车牌检测的自动化。车牌识别的素材主要基于黑白图像(灰度图)和彩色图像两大类。相对于黑白图像,彩色图像包含的信息量要大,对环境变化的适应性也强。但正如我们所知道的,车牌颜色具有多样性:蓝底白字、黄底黑字、白底黑字等各种颜色的组合,使用彩色图像处理要分析多种情况。而对于室内的车牌识别,环境差异不大,对黑白图像的分析处理也可以获得较好的效果。本文把黑白图像作为处理对象,在从RGB变换到HLS时,避开经典算法中计算量大、耗时多的弱点,采用一种快速变换的算法,较好地满足了系统的实时性要求。传统的图像分割,多是直接对原始图像进行二值化,以期区分出目标和背景。但这种方法无论对目标还是背景的图像特性依赖都很大,更容易受外界环境的影响,使得分割效果大打折扣。本文提出的车牌图像经过边缘检测后再进行二值化的方法,能够准确地把车牌从背景中分割出来,为车牌字符的定位、识别奠定基础。在识别阶段,本文采取特征提取和多层感知器网络(MLPN)相结合的识别方法,既可以有效减少信息量的输入,又充分发挥了神经网络的学习和容错能力,从而达到较高的识别精度。更多还原

【Abstract】 License Plate Recognition techniques have been widely used in modern transportation and play an active role in increasing productivity and reducing labor intensity.This study is based on the “License Plate Auto-Recognition System” project. It investigates related theories of computer vision and pattern recognition, designs a license plate recognition algorithm, and applies to real-world scenarios. The automatic license plate recognition is realized in this study.The basic license plate recognition can be classified into two categories: black and white (gray scale) image and color image. In contrast to black and white images, color images provide more information and better adaptation to surrounding environments. Multiple possible combinations of the color of license plate such as white words on blue background, black words on yellow background, black words on white background and so on necessities the analysis of various situations. However, for in-door license plate recognition, satisfactory results can be obtained by analyzing only black and white figure because of similar environments. This research focuses on black and white images, reduces the computational cost of classical algorithms for the transformation of RGB to HLS with a new fast conversion algorithm, and meets the real-time requirements of the system.Most of the classical image segmentation methods utilize binarization to differentiate object and its background. However, this method relies largely on the characteristics of both the object and background, and the quality of segmentation is easily impacted by the environment. This study proposes binarization after edge detecting, which can accurately separate license plate from its background and serve as the basis for locating and recognizing individual numbers and letters on license plates.At the recognition stage, we propose the combination of characteristic input and Multi-layer Perception Network (MLPN), which both reduces information input and <WP=5>improves self-study and fault tolerance of artificial neural network, resulting in a more precise recognition.更多还原