The proposed method is to find a document region in an image for the estimation of a homography. When we have the position information of four points in a plane, the homography could be estimated simply [2]. However, the problem is how to find these four feature points. For example, corner points of the paper of the document could be used as the feature points to rectify the image of a document, if the shape of a paper with a document is rectangle. Edge information or corner detection algorithms could be used to find these feature points Unfortunately, there are many outliers, because both characters in a document and the boundary of a paper make strong edge points. To eliminate or reduce outliers, text regions in a document are removed by decomposed edge planes according to its directional information in this paper. After removing text regions in each decomposed edge plane, the Hough transform is applied in each decomposed edge plane with reduced angle domain. This results in more robust line detection compared with using a single edge plane with text regions. A homography is estimated with the detected feature points, and the document region in an image is warped with it. This paper is organized as follows. In section II, the background and related works are explained. Section III describes the proposed system in detail. The implementation results are presented in section IV. Finally, section V concludes the paper.