Offline Handwritten Uighur Word Recognition Based on Segmentation-driven and Two-level DTW

Offline handwritten Uighur scripts is cursive and have a large vocabulary, which makes the word recognition more complicated. In this paper, we propose a segmentation-driven recognition algorithm for offline handwritten Uighur word based on grapheme analysis and two-level DTW (dynamic time wrapping). Firstly, a MSAC (main segmentation and additional clustering) algorithm is adopted to over-segment a handwritten Uighur word into two grapheme sequences. After then, a hierarchical hybrid Uighur character classifier is designed to enhance the character recognition accuracy. Finally, a novel maximum likelihood algorithm with two-level DTW is presented to select the best hypothesis of character sequence from grapheme merging and decide the word class. Experiment results show that the proposed algorithm can achieve high character segmentation accuracy and word recognition rate simultaneously.