基于主元分析的车牌图像倾斜校正新方法

为解决机动车牌图像倾斜将对其字符分割与识别带来不利的影响,提出一种基于主元分析(PCA)的车牌图像倾斜校正新方法。在该方法中,PCA被用于求取坐标变换矩阵以进行图像旋转修正。将原始的像素坐标矩阵经过中心化后转换为2维协方差矩阵,再奇值分解为能反映图像倾斜方向的2维对角矩阵和坐标变换矩阵。算法的时间复杂度分析与试验结果均表明:相对于Hough等搜索倾角的校正方法,PCA方法缩短了计算时间1￣2个数量级,并且在污迹、光照不均等条件下也能获得较好效果。

A Novel License Plate Slant Correction Method Based on Principal Component Analysis

The authors present a new method to remedy the negative effect arising from slant vehicle license plate's character segmentation and recognition based on principal component analysis(PCA).The geometrical transform matrix to correct the image is acquired by the PCA.The image data set is arranged to coordinate matrix into two-dimension covariance matrix,on which centering is operated.Then by the singular value decomposition,the matrix is refold to the bi-diagonal matrix and coordinate transform matrix,which are consistent with the main slant direction of the license image.The time complexity of the PCA method is analyzed in the paper.The experiment demonstrates that the new method raises the computation rate by 1~2 orders compared with the Hough method and is effective when dealing with images of dirty vehicle license or in variant lighting conditions.