Key points are not available for this paper at this time.
An offline recognition system for Arabic handwrittenwords is presented. The recognition system is based ona semi-continuous 1-dimensional HMM. From each binaryword image normalization parameters were estimated. Firstheight, length, and baseline skew are normalized, then featuresare collected using a sliding window approach. Thispaper presents these methods in more detail. Some parameterswere modified and the consequent effect on the recognitionresults are discussed. Significant tests were performedusing the new IFN/ENIT - database of handwritten Arabicwords. The comprehensive database consists of 26459Arabic words (Tunisian town/village names) handwrittenby 411 different writers and is free for non-commercial research.In the performed tests we achieved maximal recognitionrates of about 89% on a word level.
Pechwitz et al. (Tue,) studied this question.