基于失真图像特性测量的扫描图像校正技术

提出一种扫描图像校正技术，解决书籍扫描时图像背景灰度不匀、几何畸变和局部模糊的问题。该方法不依赖扫描设备的物理参数，完全根据对失真图像的测量来计算校正因子。选取适当的采样窗1：3，基于直方图分析导出背景灰度因子，通过内插校正几何失真，最后进行空变高通滤波以减轻字迹模糊。通过建立背景灰度因子和缩小系数的关系实现对扫描设备的标定，从而在校正同一设备输出的其它图像时可根据背景灰度的变化而不必再检测页面边缘。与原始扫描图像相比，处理后的图像质量大为改善，且光学字符识别（OCR）处理的识别率达到82．6％。

Scanned Image Correction Technique Based on the Measurement of Distorted Images

An image correction technique is proposed to remove distortion ,which is featured by background non-uniformity ,geometric distortion and local blurring near binding ,from images scanned from thick books. The method does not rely on any physical parameters of the scanner ,and the correction is performed entirely based on measuring the distorted image. An appropriate sampling window is moved across the image,and a background grey factor is obtained for correction of grey values in the whole image. Taking a curved book edge in the distorted image as reference,a distortion factor is found and used in the interpolation for geometric distortion correction. Finally a space varying high-pass convolution mask is used to sharpen the image in order to alleviate blurring. By establishing relationship between background grey factor and shrinking coefficient to calibrate the scanner, information of the page edge is not needed in correcting other images produced by the same equipment. The corrected images are suitable for human viewing,and OCR recognition ratio is up to 82.6%.