Off-Line Arabic Handwritten Word Segmentation Using Rotational Invariant Segments Features

Shubair Abdulla¹, Amer Al-Nassiri², and Rosalina Abdul Salam³

¹Faculty of Education and Basic Sciences, Ajman University of Science and Technology, UAE

²Faculty of IT, Ajman University of Science and Technology, UAE

³School of Computer Sciences, University SaMalaysia, Malaysia

Abstract: This paper describes a new segmentation algorithm for handwritten Arabic characters using Rotational Invariant Segments Features (RISF). The algorithm evaluates a large set of curved segments or strokes through the image of the input Arabic word or subword using a dynamic feature extraction technique then nominates a small “optimal” subset of cuts for segmentation. All the directions of stroke are converted to two main segments: '+' and w'-' RISF. A list of nominated segmentation points are prepared from the '+' segments and evaluated according to special conditions to locate the final segmentation points. The RISF algorithm was tested by using our new designed database AHD/AUST and the IFN/ENIT database. It has achieved a high segmentation rate of 95.66% on AHD/AUST and 90.58% on IFN/ENIT handwritten Arabic databases.

Keywords:  Feature extraction, Arabic character segmentation, cursive writing, Arabic words database.

Received June 24, 2006; accepted March 19, 2007