Abstract:

As an important writing material, the interpretation of Chu Jian and silk has always been the key research direction of ancient philology. However, at present, the Chu system of bamboo and silk text interpretation mainly relies on artificial means to analyze the single character form, and there is a lack of computer vision technology for font recognition of massive text plates. Aiming at the difficulty of image recognition of a large number of Chu script and silk text, this paper proposes an integrated learning strategy based on image classification method for Chu script and silk text, which is not limited to the microscopic perspective of single deep neural network model and single text image analysis, combined with the inherent characteristics of Chu script and silk text. Different deep learning networks were used to extract the common morphological features of Chu Jian and silk text images, and the final classification results were obtained in the form of voting, and a technical framework for automatic and efficient recognition of massive Chu Jian and silk text images was constructed. The framework is applied to recognize the text images in some unearthed silk materials with an accuracy of 96.72% ,which fully proves the feasibility and effectiveness of the framework and provides a new way for the study of ancient Chinese characters.

Key words: