Abstarct: Document images produced by scanner or digital camera, usually suffer from geometric distortions and photometric distortions. Both of them deteriorate the performance of OCR systems. In this thesis harmonic inpainting is used to remove photometric distortion. To improve accuracy of this algorithm, we propose that diagonal neighbors can be used too. This will lead to an improvement in background intensity estimation. Although, to reduce computational complexity, we propose that in implementation of inpainting algorithm, first we must use main quad neighbors and then the diagonal neighbors. Using this idea, computational complexity of proposed method won’t be increased. Also, Gaussian smoothing and Moving Average filters are used to improve quality in background estimation. Although, in order to correct Geometric distortion, after extracting lines from image, each line is divided into some vertical strips. In each strip, coordinates of the point which its horizontal projection is maximum, is stored. Curvature of each line is determined by third order function using this set of points. At the end, geometric distortion of each line is corrected using Perspective transform estimation. Also, to determine curvature of short lines, curvature of previous or next long lines are used. The proposed method was implemented on Persian and English databaxses and has been compared with other methods. The obtained results indicate that proposed method is powerful and precise in eliminating both types of distortions.