基于低秩纹理变换不变性的车牌倾斜校正方法

Vehicle license plate tilt correction approach based on transform invariant low-rank textures

文章提出一种适用范围更广的基于低秩纹理变换不变性的车牌倾斜校正方法。首先将车牌图像看成一个矩阵,然后利用对称规则结构的低秩特性把经过旋转、仿射等变换的图像恢复成低秩纹理。根据不同的环境条件,对随机拍摄的倾斜车牌图像进行校正,实验表明即使面对一些如无边框、车牌污染、环境光遮蔽、噪声大等恶劣条件,采用所提出的方法也得到了很好的校正结果。文中提出了一种借助于图像整体信息进行倾斜校正的新思路,该方法摒弃了以往考虑角点、边缘等局部特征的思路,具有更高的精确度和鲁棒性。实验结果证实所提出的方法是有效的,且满足整个识别系统的实时要求。

A vehicle license plate tilt correction approach based on transform invariant low-rank textures is proposed, which has a wider range of applications. Firstly, considering the license plate image as a matrix, the low-rank characteristic of regular structures and symmetric patterns is utilized to restore the image undergoing rotation and affine transformation into a low-rank textures. According to different environmental conditions, the tilt correction for license plate images which are randomly shot is carried out. Experimental results show that even in the face of some harsh conditions, such as borderlessness, pollution, ambient occlusion and image noise, excellent correction results are obtained by the proposed method. A new idea of tilt correction by means of the image overall information is put forward, which abandons the previous point of view from the corner, edge and other local features, and therefore it has a higher robustness and accuracy. Experimental results demonstrate that the proposed method is effective and can satisfy the demands of the whole vehicle license plate recognition system in real time.