Arabic Font Recognition using Decision Trees Built from Common Words

Abstract

We present an algorithm for a priori Arabic optical Font Recognition (AFR). The basic idea is to recognize fonts of some common Arabic words. Once these fonts are known, they can be generalized to lines, paragraphs, or neighbor non-common words since these components of a textual material almost have the same font. A decision tree is our approach to recognize Arabic fonts. A set of 48 features is used to learn the tree. These features include horizontal projections, Walsh coefficients, invariant moments, and geometrical attributes. A set of 36 fonts is investigated. The overall success rate is 90.8%. Some fonts show 100% success rate. The average time required to recognize the word font is approximately 0.30 seconds.