Open-Set Text Recognition via Character-Context Decoupling
Chang Liu, Chun Yang, Xu-Cheng Yin
Abstract
The open-set text recognition task is an emerging chal-lenge that requires an extra capability to cognize novel characters during evaluation. We argue that a major cause of the limited performance for current methods is the con-founding effect of contextual information over the visual information of individual characters. Under open-set sce-narios, the intractable bias in contextual information can be passed down to visual information, consequently im-pairing the classification performance. In this paper, a Character-Context Decoupling framework is proposed to alleviate this problem by separating contextual information and character-visual information. Contextual information can be decomposed into temporal information and lin-guistic information. Here, temporal information that mod-els character order and word length is isolated with a de-tached temporal attention module. Linguistic information that models n- gram and other linguistic statistics is sepa-rated with a decoupled context anchor mechanism. A va-riety of quantitative and qualitative experiments show that our method achieves promising performance on open-set, zero-shot, and close-set text recognition datasets.