Hmm-Based System for Recognizing Words in Historical Arabic Manuscript

This paper presents an omni-font Arabic word recognition system. The system is based on multiple Hidden Markov Models (HMMs). Each word in the lexicon is represented with a distinct HMM. The proposed system first extracts a set of spectral features from word images, then uses those features to tune HMM parameters. The performance of the proposed system is assessed using a corpus that includes both handwritten and computer-generated scripts. The likelihood probability of the input pattern is calculated against each word model and the pattern is assigned to the model with the highest probability.