Correction of perspective text image based on gradient method

This paper presents an approach to address the problem of the correction of perspective distortion text image which every vertex has maximum or minimum vertical or horizontal coordinate in one fourth image. In order to recognize this kind of perspective distortion text image by OCR (Optical Character Recognition), we correct it into a hardly distorted regular text image based on a gradient method. Our approach is binarizing and denoising the distortion text image firstly. Then a gradient method is applied to obtain the four vertexes of the irregular convex quadrilateral composed from the perspective distortion text block. Thus, the mapping parameters of the correction model can be obtained through one-to-one relation between the irregular quadrilateral and the rectangle of the original text image. At last, the perspective distortion text image is corrected with the perspective transformation model. The experimental results show that our approach can correct the kind of perspective text image effectively.