A Chinese handwriting word segmentation method via faster R-CNN

Abstract

The segmentation of Chinese handwritten document image into individual words is an essential step for the character recognition. Conventional methods frequently use feature extraction and classification algorithm to segment. However, since the features of the words mostly depend on people, it is considered a difficult task. In order to avoid this problem, we use a method of object detection—Faster R-CNN. The words are treated as the especial object and people do not concern on features extraction. Experimental results on HIT-MW databases show that our method achieves the preferable performance.