Skew Correction of Textural Documents

AbstractTwo algorithms for accurate skew detection and correction of textual documents are presented. They depend on finding a horizontal RLSA image of the skewed document. The average skew of selected black connected components in the RLSA image is considered as the skew angle for the whole document which is finally rotated in the opposite direction by that amount to obtain the final corrected image. The first algorithm is a single-pass skew detection and correction algorithm, i.e., the input image is either not rotated or rotated only once. The second algorithm is a multi-pass skew detection and correction algorithm, i.e., the input image is either not rotated or rotated at least once. A total of 50 pages were used to test the algorithms. The maximum absolute measured error after correction is 1.53° and 0.36° for the first and second algorithms, respectively