基于曲量场空间的车牌定位与识别

Vehicle License Plate Locating and Identifying Based on the Curved Field Space

提出了一种新的采用曲量场空间描述的车牌定位与识别算法.该算法首先在彩色车牌图像中进行亮度扫描,然后拟合渐变梯度向量,并以曲量子进行点阵描绘并消除彩色信息,最后以采样的曲量子群分别融合成车身、车牌及字符的曲量子空间,将融合的车牌、字符曲量子空间进行边缘曲量子的光滑衔接组成曲量场并提取场内曲量的空间数据及其曲量约束信息进行有损字符的修正,识别出车牌信息.该算法抓住了车牌的互异彩色域特征,继而将互异彩色域特征采用具有空间连续性规律约束的曲量场进行描述识别.大量实验表明,该算法克服了特征域旋转、光照强度变化以及多角度有损及模糊对车牌定位与识别的影响,具有速度快,识别准确,适应性强的特点.

A novel approach to license plate locating and identifying based on the Curved Field Space is proposed in this pa- per. Firstly, by detecting the brightness information in the color vehicle image, the gradual gradient vectors are fitted, and then the dot matrix depiction with curved quantum is used to describe the gradual gradient vectors and the color information is eliminated. Then, the sampling curved quantum groups are merged into sub-curved spaces which include car＇s bcxty, plate and chars. The sub- curved spaces of plate and chars which are formed of curved quantum groups constitute the ctu＇ved field space by linking up with each edge curved quantum smoothly. Finally, recognition results are obtained by depth and dimension both of which are extracted from Curved Field Space. Experiments show that the proposed approach has overcame the problem of feature rotating, varied angle, wear and fuzzy chars, and achieves good locating and identifying results.