基于小波变换和神经网络的车牌识别系统研究

【作者】 所玛；

【导师】 张忠；

【作者基本信息】 黑龙江大学 ， 信号与信息处理， 2009， 硕士

【摘要】 本文将现代信号处理技术——小波技术,应用到智能交通领域,即车牌识别技术当中来,并结合改进的BP神经网络对车牌字符进行分类识别。主要的研究内容和成果如下:首先,对车辆图像进行灰度变换、平滑处理,通过分析传统图像增强的方法,提出了一种基于小波变换的高频非线性增强算法;结合边缘检测、形态学处理、投影法等算法,提出了一种基于小波变换的车牌定位算法。其次,对车牌图像进行二值化、倾斜角校正处理,提出了一种基于小波变换的局部自适应多阈值消噪方法;在对字符进行分割的过程中,先去除车牌区域的上下边框,利用垂直投影和先验知识相结合的方法初步分割出单个字符区域,再寻找字符的形态学连通域,切出字符矩形最小区域,对误切分汉字进行合并处理。最后,对字符进行归一化、细化处理;采用分类识别的方法,用改进的BP神经网络对字符进行识别。更多还原

【Abstract】 In this dissertation, modern signal processing technique, namely wavelet technology is applied to the intelligent transportation field, which is License Plate Recognition System. And improved BP neural network is adopted to recognize license plate characters. The main contents and achievements are as follows:Firstly, the car images are processed by gray transformation and smooth filtering. By way of analyzing the methods of traditional image enhancement, we put forward an algorithm of high-frequency nonlinear enhancement based on the wavelet transformation. Combining with algorithms of edge detection and morphology processing and projection method, we put forward an algorithm of license plate location based on the wavelet transformation.Secondly, the license plate images are processed by binary method and correcting inclination angle. We put forward an algorithm of local adaptive threshold removing noise based on the wavelet transformation. During the process of character segmentation, we remove the margin height of the license plate area, use vertical projection information and the prior knowledge to get single characters, look for the morphology connected domain of the character, cut off the rectangle minimal area, and combine the error segmentations.Finally, we normalize the character images and make them thinner; By way of classified recognition, we use improved BP neural network to recognize license plate characters.更多还原