基于机器学习的模糊车牌的识别.pdf

武汉科技大学 硕士学位论文 第 I 页
摘 要
智能交通系统作为分析处理城市和公路上的车辆的工具被迅速地发展。车牌识别系统
涉及到计算机视觉、图像处理、模式识别等多个学科。由于应用场景及车牌本身的复杂性，
该系统需解决的问题相当复杂，比如外界光照条件恶劣、特殊天气条件、复杂的非车牌区
域干扰等因素都会影响车牌识别系统的性能。本文选用抗噪性较好的小波包变换对图像进
行增强；根据车牌图像具有 Haar 特征采用 Adaboost 算法训练数据区分含有车牌的图像和
不含车牌的图像，采用 SUSAN 角点检测方法标记所有角点位置，然后定义一个矩形滑动
窗口扫描整幅图像，最后滑动窗口中白色像素点最多的位置即为车牌区域，从而定位出车
牌区域。根据 Hough 变换进行水平分割，之后再进行垂直分割准确的分割出各个字符，最
后采用 BP 神经网络的方法是识别出分割出来的字符。

关键词：车牌定位；Adaboost 算法；SUSAN 角点；字符分割；BP 神经网络；字符识别
第 II 页 武汉科技大学 硕士学位论文
Abstract
As analytical tools Intelligent Transportation Systems dealing with the city and highway
vehicles are developing rapidly. License Plate Recognition System related to computer vision,
image processing, pattern recognition, and other disciplines. Scenarios and the license plate itself,
the complexity of the issues to be addressed in the system is quite complex, such as outside
lighting conditions, special weather conditions, the complexity of the non-plate region
interference factors can affect the performance of the license plate recognition system. The
selection of anti-noise better wavelet packet transform image enhancement; according to the
license plate image with Haar features using Adaboost algorithm training data to distinguish
between images with and without license plates containing the license plate image, using the
SUSAN corner detection method to mark all corner location, and then define a rectangular
sliding window to scan the entire image, the location of the white pixels in the last sliding
window is the plate area to locate the plate region. Vertical split level split according to the
Houg