基于形状上下文的低质量车牌字符识别方法

摘要

车牌字符识别的问题已经被广泛研究,但是对于一些低质量字符的识别率还是有待提高,特别是形近的字符,如(4,A).针对这一问题提出一种结合局部特征和形状上下文特征的两级识别方法.在第一级分类时(称为粗分类),将形近字符识别为一类.若粗分类结果不是形近字符,则输出识别结果；否则进入第二级分类器进行细分类.细分类使用形状上下文的方法,并结合显著特征区域的思想,针对不同类别的形近字符,运用基于显著特征区域的形状上下文方法进行特征提取,再通过图像匹配得到识别结果.实验证明这种方法能够有效地对低质量形近字符进行识别.%The problem of license plate characters recognition has been investigated extensively.However the recognition rate in regard to poor-quality characters is to be improved still,in particular those characters in very similar shapes,for example,4 and A.To address this issue,we propose a two-stage recognition approach which combines the local feature with the shape context feature of the characters.In first-stage classifier (rough classification),the nearly-shaped characters are recognised as the same category.If the results of rough classification are not the nearly-shaped characters,then output them,otherwise get into second-stage classifier for fine classification.The fine classification uses the method of shape context and combines the idea of remarkable feature region,in light of the nearly-shaped characters with different categories,it utilise the remarkable feature region-based shape context method to extract the features,and to obtain the recognition result through image matching.Experiment proves that this method can effectively recognise the nearly-shaped characters with poor quality.