A skew detection and correction technique for Arabic script text-line based on subwords bounding

Abstract

Text-line skew detection and correction is the first step in Arabic document recognition and analysis. It is a crucial pre-processing stage of Arabic Character Recognition (ACR). It has a direct effect on the dependability and efficiency of other system stages such as baseline detection, segmentation and feature extraction stages. In this paper an efficient skew detection and correction method for Arabic handwritten text-line based on sub-words bounding is presented. It is constructed from three stages including: pre-processing, skew detection and skew correction stages. The proposed method estimates a text-line baseline based on calculating the middle point for its sub-words bounding. Then align the text-line components on the estimated baseline. The proposed method is implemented on 3960 text-line handwritten images, which were written by 40 writers. It is discussed with the horizontal projection method in terms of effectiveness. The proposed method obtained an accuracy ratio of 96.15%, and takes 6.7 seconds as average time. It can also automatically detect text baselines of documents with any orientation.