Radical analysis network for learning hierarchies of Chinese characters. (July 2020)

• Record Type: Journal Article
• Authors: Zhang, Jianshu; Du, Jun; Dai, Lirong
• Is Part Of: Pattern recognition. Volume 103(2020:Jul.)
• Publication Date: 2020-07
• Subjects: Radical -- Attention -- Chinese character -- Few-/zero-shot learning; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2020.107305
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: A novel radical-based Chinese character/text recognition method is proposed. The proposed method has the capability of few-/zero-shot learning. The hierarchical radical structure of 27, 533 Chinese characters is released. Achieved the first place on ICPR MTWI 2018 challenge. Abstract: Chinese characters have a valuable property, this is, numerous Chinese characters are composed of a compact set of fundamental and structural radicals. This paper introduces a radical analysis network (RAN) that makes full use of this valuable property to implement radical-based Chinese character recognition. The proposed RAN employs an attention mechanism to extract radicals from Chinese characters and to detect spatial structures among the radicals. Then, the decoder in RAN generates a hierarchical composition of Chinese characters based on the knowledge of the extracted radicals and their internal structures. The method of treating a Chinese character as a composition of radicals rather than as a single character category is a human-like method that can reduce the size of the vocabulary, ignore redundant information among similar characters and enable the system to recognize unseen Chinese character categories, i.e., zero-shot learning. Through experiments, we assess the practicality of RAN for recognizing Chinese characters in natural scenes. Furthermore, a RAN framework can be proposed for scene text recognition with the extension of a dense recurrent neural network (denseRNN) encoder, a multihead coverage attention model and HSV representations. The proposed approach achieved the best performance in the ICPR MTWI 2018 competition.