The Effect of Fonts Design Features on OCR for Latin and Arabic

Abstract

Huge digitized book projects have been launched recently like Google Book Search Library project and MLibrary. They basically depend on optical character recognition systems (OCR) to convert scanned books or documents into editable and text searchable e-books. Although research on OCR area pursued over decades, very few of them focus on the effect of typeface design on the recognition rate. We took a further step by conducting systematically two observational experiments on Latin and Arabic typefaces using OCR tools. A collection of 18 Latin and 13 Arabic typefaces have been tested in two sizes using six OCR packages in total. In addition, confusion tables have been constructed to show similarities among some characters of the alphabet. Extensive analyses were made to find correlations between the recognition rates and font design characteristics. Our findings indicate that some font design features showed influence and negative effects on the recognition rate. This will guide typeface designers to produce recognizable typefaces and publishers to select the appropriate recognizable fonts.