Freeman Arabic Classification Tree (FACT) For Arabic Character Recognition System

A. AL-Nassiri


Image enhancement, Arabic character recognition, pattern recognition and document analysis, text recognition.


This paper describes the design and implementation of a system that classifies and recognizes machine-printed Arabic characters with prior segmentation using a new suggested simple classification tree called Freeman Arabic classification tree (FACT) .The technique is based on describing symbols in term of shape primitives derived using Freeman chain algorithm (Fives Paired Freeman Chain FPFC)[1]. A heuristic segmentation algorithm is initially used to segment each word .This algorithm is based on describing Arabic words and sub words in term of direction primitives using Freeman chain technique to obtain an enhanced fives pair freeman chain as a feature for each isolated character[1].At the recognition time, the FPFC primitives ( for characters, holes, and dots)or features are used in matching stage. The detected FPFC , for both Arabic characters under recognition and prototypes, are classified according to the suggested classification tree (FACT). In this FACT, the Arabic characters are classified into groups according to the presence of hole, dots, and FPFC features. The advantage of using FACT versus lookup table approach used in [1] , is that the result of recognition is optimized with regard to the FACT approach. This approach deals with the problem of dot and holes related to Arabic characters in completely different way than that used in [1].Results of preliminary experiments using this approach shows a recognition rate of 98% for noise free Arabic text and 88% for scanned Arabic text based on 2200 Arabic words.

Important Links:

Go Back