Recognition of Handwritten Chinese Text by Segmentation: A Segment-Annotation-Free Approach

Online and offline handwritten Chinese text recognition (HTCR) has been studied for decades. Early methods adopted oversegmentation-based strategies but suffered from low speed, insufficient accuracy, high cost of character segmentation annotations. Recently, segmentation-free based on connectionist temporal classification (CTC) attention mechanism, have dominated the field HCTR. However, people actually read by character, especially ideograms such as Chinese. This raises question: are really best solution to HCTR? To explore this issue, we propose a new segmentation-based method recognizing that is implemented using simple yet efficient fully convolutional network. A novel weakly supervised learning proposed enable network be trained only transcript annotations; thus, expensive annotations required previous can avoided. Owing lack context modeling in networks, contextual regularization integrate information into during training stage, which further improve performance. Extensive experiments conducted four widely used benchmarks, namely CASIA-HWDB, CASIA-OLHWDB, ICDAR2013, SCUT-HCCDoc, show our significantly surpasses existing both online HCTR, exhibits considerably higher inference speed than CTC/attention-based approaches.