Abstract: Image normalization is very important in quality inspection. Aiming at the distortion of printed images, an improved method of normalization is proposed for perspective transformation. Firstly, the problem of the threshold adaptive ability is improved for traditional Canny detection to obtain accurate linear edge information of image. The threshold parameters are obtained by iterative method, and the parallel straight line is filtered out effectively, which is too close to image edge. Image is divided into three categories to get the feature point information of the image, and the perspective transformation matrix is used to normalize the image. Experimental results show that relative inaccuracy of the proposed algorithm is reduced by 41.2% for printed images, which can solve image distortion problem effectively.

Key words: normalization, printed matter, transmission transformation, Canny edge detection

摘要： 印刷品图像归一化在其质量检验中是极其重要的。针对获取的印刷品图像存在一定程度的畸变，提出一种改进的透视变换的归一化方法。为了准确获取图像的直线边缘信息，改进传统Canny检测中阈值自适应能力差的问题，用迭代法获取阈值参数，并加入滤波器有效滤除过于贴近图像边缘的平行直线；针对传统透视变换特征点无法获取问题，首次提出将图像分为三类状态下分别获取图像的特征点信息，求解得出透视变换矩阵对图像进行归一化。实验结果表明，对采集的印刷品图像该算法相对不准率下降了41.2%，可有效解决图像存在的畸变问题。

关键词: 归一化, 印刷品, 透射变换, Canny边缘检测