基于空间分布的最大类间方差牌照图像二值化算法

车辆牌照识别VLPR(Vehicle License Plate Recognition)是智能交通系统ITS(Intelligent Transportation System)的重要部分，有着巨大的社会和经济效益。车辆牌照图像二值化方法的研究是VLPR中的关键技术，通常都采用经典的Bernsen算法和Otsu算法．但由于光照不均、摄像机畸变、曝光不足、动态范围太窄和车辆牌照被污染等原因，车辆牌照图像的质量往往不佳，常存在严重伪影和字符边缘模糊，极大地影响了牌照图像二值化效果，Bernsen算法和Otsu算法也不能很好地克服上述问题。为此，提出了一种新的牌照图像二值化算法CASDA(Cluster Algorithm Based on Spatial Distribution Analysis)，把牌照图像空间分布特性与最大类间方差的统计特性有机结合，大大加快了最大类间方差模式分类速度．与Bernsen、Otsu、LEVBB等算法比较，CASDA具有速度快，能消除不均匀光照引起的伪影，极少出现笔划断裂等优点，二值化效果好．

License plate binarization algorithm based on analysis of the spatial distribution and maximum variance between clusters

VLPR (Vehicle license plate recognition) is an important part of ITS (Intelligent Transportation System), which has great social and economic values. Vehicle license plate binarization algorithms comprise a critical technology in VLPR. Generally, Bernsen algorithm and Otsu algorithm are used in vehicle license plate binarization. However, because of the deficiency in illumination, vidicon, exposure, dynamic range and license plate pollution, the quality of vehicle license plate images is always very poor, with serious false shadows and blur character edges existing on these images. Bernsen algorithm and Otsu algorithm connet overcome these problems, so a new license plate binarization algorithm\|CASDA (Cluster Algorithm Based on Spatial Distribution Analysis) is presented based on maximum variance between clusters. CASDA algorithm has the advantage of short computing time, few broken strokes, robust to subnormal illumination and noise compared with the famous algorithm Bernsen, Otsu and LEVBB, has batter performance.