Arabic Text Recognition System

Andrew Gillies, Erik Erlandson, John Trenkle, Steve SchlosserNonlinear Dynamics Incorporated123 N. Ashley Street, Suite 120Ann Arbor, MI 48104AbstractThis paper describes a system for the recognition of Arabic text in document images.The system is designed to perform well on low resolution and low quality documentimages. On a set of 138 page images digitized at 200x200 dpi the system achieved a93% correct character recognition rate. On the same pages digitized at 100x200 dpi, thesystem achieved an 89% character recognition rate. The systems processes a typicalpage with simple layout and 45 lines of text in 90 seconds on a 400 Mhz Pentium IIrunning Linux.