一种基于MSER和SWT的新型车牌检测识别方法研究

为了降低拍摄距离、光照等对车牌识别的影响,提高复杂背景下车牌识别准确率,提出了一种基于最大极值稳定区域(maximally stable extremal regions, MSER)和笔画宽度变换(stroke width transform, SWT)的新型车牌检测识别方法。该方法首先进行MSER提取和Canny边缘检测,并根据车牌字符特征对二者相与运算后的MSER筛选;然后在筛选后的区域内做基于形态学处理的SWT和SW筛选,聚合筛选后区域,结合车牌几何特征完成车牌精定位;最后校正分割定位成功区域内连通域,提取骨架并归一化,与细化和归一化后的模板匹配。利用HU不变矩和网格特征识别首字符汉字,采用扫描跳跃点统计编码识别数字和字母。实验结果表明:该方法定位准确率高达94.86%,识别准确率达96.14%。该方法对远距离、变光照获取的复杂背景下,车牌检测识别具有较高的准确率和鲁棒性。

A New Vehicle Licence Plate Recognition Method Based on MSER and SWT

In order to reduce the influence of camera distance and the light on license plate recognition, improving the recognition accuracy of license plate with complex background, a new license plate recognition method based on maximally stable extremal regions (MSER)and stroke width transform(SWT) has been proposed. Firstly, the MSER extraction and Canny edge detection are carried out and the filter of the MSER based on geometric features of license plate character is conducted, then the SWT and SW selection of the selected areas is conducted based on morphological processing, the selected regions are gathered and the license plate is located combined with their features. Finally, the segmentation of the location region is checked and its skeleton are extracted and normalized, then, it is matched with the thinned and normalized template. The first character Chinese characters are recognized by HU invariant moments and grid features, and the digits and letters are identified by scanning skip point statistics code. Experiment results show that the accuracy of the location is as high as 94.86% and the accuracyof the recognition is as high as 96.92%, which indicates that this method has high accuracy and robustness for license plate detection and recognition in complex background with long distance and variable illumination.