A HMM-Based Arabic/Latin Handwritten/Printed Identification System

For document analysis and recognition systems, script identification is considered as an important preprocessing step in the design of multi-scripts OCR system. In this paper, we propose a novel HMM based identification system to recognize on only one level the writing type (handwritten or machine-printed) and the script nature (Arabic or Latin) of the input image. The proposed system is based on Histogram of Oriented Gradient (HOG) features which have demonstrated an interesting properties for script characterization. Experiments have been conducted on word and line images collected from public databases and show promising results.