Knowing varieties of writing a letter in a word or a subword in different handwriting styles is very beneficial in recognition specifically for online recognition. In this paper, TMU-OFS dataset consisting of 1000 frequent Farsi subwords is employed to study Farsi handwriting styles. The subwords are grouped based on their delayed strokes and their main bodies, separately. The handwriting styles in this dataset are analyzed and the wrongly spelled or incorrect structural samples are extracted. Finally, the second version of the dataset is introduced by considering the handwriting styles. The preliminarily results show a significant improvement in recognition of subwords based on their styles.
展开▼
