Arabic handwriting recognition using structural and syntactic pattern attributes

In this paper, we present research results on off-line Arabic handwriting recognition using structural techniques. Statistical methods have been more common in the reported research on Arabic handwriting recognition. Structural methods have remained largely unexplored in this regard. However, both statistical and structural techniques can be effectively integrated in multi-classifier based systems. This paper presents, to our knowledge, the first integrated offline Arabic handwritten text recognition system based on structural techniques. In implementing the system, several novel algorithms and techniques for structural recognition of Arabic handwriting are introduced. An Arabic text line is segmented into words/sub-words and dots are extracted. An adaptive slant correction algorithm that is able to correct the different slant angles of the different components of a text line is presented. A novel segmentation algorithm, which is integrated into the recognition phase, is designed based on the nature of Arabic writing and utilizes a polygonal approximation algorithm. This is followed by Arabic character modeling by 'fuzzy' polygons and later recognized using a novel fuzzy polygon matching algorithm. Dynamic programming is used to select best hypotheses of a sequence of recognized characters for each word/sub-word. In addition, several other key ideas, namely prototype selection using set-medians, lexicon reduction using dot-descriptors etc. are utilized to design a robust handwriting recognition system. Results are reported on the benchmarking IfN/ENIT database of Tunisian city names which indicate the robustness and the effectiveness of our system. The recognition rates are comparable to multi-classifier implementations and better than single classifier systems.