基于最大字符像素距离计算的牌照属性判别方法

A Method about License Plate Attribute Identification Using Maximum Distance of Character Pixels

基于"九二"式牌照识别系统不适合识别"二○○二"式牌照的现状,提出了一种针对目前现有牌照标准各异情况下判别牌照属性(本文中属性意指单行牌照或双行牌照)的新方法。该新方法不同于以往方法的两个步骤是图像修剪和最大字符像素距离的计算,它克服了传统方法中仅仅依靠字符投影来进行牌照属性判别的弊端。该方法是基于"二○○二"式牌照与"九二"式牌照在尺寸、字符数目及其排列方式等上的差异,着眼于字符的空间分布特点,采用图像修剪、留白检测、最大字符像素距离计算和基于不同优先级的分布式信息融合等几个步骤来进行牌照属性判别。试验结果表明,该方法判断准确,鲁棒性强,是对现有牌照识别系统的有益扩展和补充。

A novel algorithm for the license plate attribute identification is presented （attribute means one-row or two-row characters that license plate really has in this paper）. The proposed approach provides a solution to the traditional vehicle license plate recognition system to differentiate the existing license plates with different style based on their dissimilarities in size, character number and alignment fashion. It overcomes the defects caused by the traditional methods based on pixels projection. What it remarkably differs from the traditional methods is the application of image pruning and the calculation of maximum character pixels distance, which increases the identification rate to an extent. The approach takes the pixels＇ spatial distribution into account. There are four steps to perform the attribute identification, that is, object region extraction, margin detection, maximum character pixels distance calculation and hierarchical-based distributed fusion. Experimental results demonstrate the method＇s accuracy, feasibility and robustness. It would be a potentially significant contribution to the active area of license plate recognition system.