Online handwritten Tibetan syllable recognition based on component segmentation method

Syllable-based input is more preferable than character-based input for Tibetan people due to the inherent characteristics of Tibetan characters. This paper presents a component segmentation-based recognition method for online handwritten Tibetan syllables. The input syllable is over-segmented into a sequence of sub-structure blocks (stroke blocks) with two-layer segmentation point annotation using horizontal-vertical over-segmentation method. Segmentation hypotheses based on sub-structure block sequences are evaluated by fusing multiple contexts into a principled Tibetan syllable recognition framework. Component-based and character-based bi-gram models are used to represent linguistic contexts. The optimal path is searched to give the component segmentation and syllable recognition results. We evaluated the recognition performance on online handwritten Tibetan syllable database with 827 classes. Experimental results show the effectiveness of the proposed method. Our method achieved the syllable-level recognition rate of 81.23%, and is superior to character segmentation and segmentation-free methods.