ICDAR 2009 Handwritten Farsi/Arabic Character Recognition Competition

Abstract

In recent years, the recognition of Farsi and Arabic handwriting is drawing increasing attention. This paper describes the result of the ICDAR 2009 competition for handwritten Farsi/Arabic character recognition. To evaluate the submitted systems, we used large datasets containing both binary and gray-scale images. Many different groups downloaded the training sets; however, finally 4 systems successfully participated in the competition. The systems were tested on two known databases and one unknown dataset. Due to the similarity between some digits and characters in Farsi and Arabic, each recognizer was tested for digit and character sets separately. For benchmarking, only the recognition rates, as the most important characteristic, are considered. Since participants used different software and even operating systems, the relative recognition speed is not compared in this competition.