Hough Transform OCR Image Slant Correction Method
Houhg变换OCR图象倾斜矫正方法

In process of OCR image scanning input, images may slant somehow. Severe image slant makes character segmentation next step difficult and lowers overall recognition accuracy of the system. Normally, in order to improve recognition accuracy and save image rescanning time, software correction is used to correct the slant before next step recognition processes of OCR are executing. This paper presents a novel slant correction method called Hough Transform OCR Image Slant Correction Method. The method uses Hough Transform to estimate the slant of scanning images. Because of heavy computing burdens of Hough Transform, the method employs Image Pyramid Processing to overcome the defeat of Hough Transform. Horizontal edge extracting processing is also used to speed the estimating process and to improve estimating accuracy. A great many artificial images and raw scanning images are used to test performances of the methodology. Experiment results show that the method can measure out slanting degree of scanning image efficiently and accurately just in a few seconds, and bears high noise endurance and application adaptability.