Efficient Algorithms for Text Lines and Words Segmentation for Recognition of Arabic Handwritten Script

Abstract

A new methodology for Arabic handwritten document images segmentation is done in this paper to segment the documents into distinct entities as words and text lines. Based on features of Arabic scripts, the document images are divided into three main subsets of connected components where the Hough transform method is applied to them to achieve text lines segmentation. To enhance the result by avoiding the Hough transform text line detection failure, the authors used a method in postprocessing stage based on skeletonization that covers the possible false correction alarms to create proficiency vertical connected characters’ segmentation. The segmentation of the Arabic words is pointed as a two-class problem. The authors used fusion of convex and Euclidean distance metrics to calculate the distance between neighboring overlapped components, which in the Gaussian mixture modeling framework is classified as a distance of an intra-word or as an inter-word. The proposed method performance is depended on a constant and particular evaluation method that appropriate measures of the performance used to compare the segmentation of our result against the other strong researcher result. The proposed method showed higher efficiency and accuracy in the experimentation, which was conducted on two various Arabic handwriting datasets that are IFN/ENIT and AHDB.