AT-ST: Self-training Adaptation Strategy for OCR in Domains with Limited Transcriptions

This paper addresses text recognition for domains with limited manual annotations by a simple self-training strategy. Our approach should reduce human annotation effort when target domain data is plentiful, such as transcribing collection of single person's correspondence or large manuscript. We propose to train seed system on scale from related mixed available annotated the domain. The transcribes unannotated which then used better system. study several confidence measures and eventually decide use posterior probability transcription selection. Additionally, we augment using an aggressive masking scheme. By self-training, achieve up 55 % reduction in character error rate handwritten 38 printed data. augmentation itself reduces about 10 its effect pronounced case difficult