A Tree-Structure Analysis Network on Handwritten Chinese Character Error Correction

Abstract

Existing researches on handwritten Chinese characters are mainly based on recognition network designed to solve the complex structure and numerous amount characteristics of Chinese characters. In this paper, we investigate Chinese characters from the perspective of error correction, which is to diagnose a handwritten character to be right or wrong and provide a feedback on error analysis. For this handwritten Chinese character error correction task, we define a benchmark by unifying both the evaluation metrics and data splits for the first time. Then we design a diagnosis system that includes decomposition, judgement and correction stages. Specifically, a novel tree-structure analysis network (TAN) is proposed to model a Chinese character as a tree layout. Using the predicted tree layout for judgement, if the character is wrongly written, correction operation is needed for error analysis. The correction stage mainly consists of three steps: fetch the ideal character, correct the errors and locate the errors. Computing the distance between the model output and all candidate characters can generate ideal characters. With the decomposition ability of TAN, edit distance computing between predicted tree and ideal tree can lead to correction operations. By using attention visualization, we can even provide the approximate error location. Through quantitative analysis, we prove that TAN captures more accurate spatial position information than regular encoder-decoder models. Additionally, we propose a novel bucketing mining strategy to apply triplet loss at radical level to alleviate feature dispersion. Experiments on handwritten character dataset demonstrate that our proposed TAN shows great superiority on all three metrics comparing with other state-of-the-art models.