Deep learning-based oracle bone character part detection and recognition method

Abstract

The invention discloses a deep learning-based oracle bone character part detection and recognition method, and mainly relates to the technical field of oracle bone character recognition. The method comprises the following steps: S1, scanning or photographing an oracle bone rubbing to obtain an oracle bone image; S2, screening an oracle single part region in the oracle image by adopting a maximum extreme value stable region algorithm; S3, carrying out ellipse fitting on the selected oracle single part area; S4, adopting a non-maximum suppression algorithm to remove redundant repeated or wrong detection results of the same oracle single part, oracle single part detection is completed, and oracle single part data is obtained; and S5: utilizing the BN-Lenet network model to identify the oraclesingle part data. The method has the advantages that high recognition precision can be obtained, oracle bone texts are regarded as partial combinations instead of whole-character recognition, unexamined oracle bone text new characters, namely zero-sample learning, can be recognized, and the method has important application significance in oracle bone text research.

Description

technical field [0001] The invention relates to the technical field of oracle bone inscription recognition, in particular to a method for detecting and identifying oracle bone inscription radicals based on deep learning. Background technique [0002] Oracle bone inscriptions are the earliest mature writing system discovered in China so far. They are the source of Chinese characters and the root of Chinese excellent traditional culture. However, the interpretation of oracle bone inscriptions is still difficult. There are few studies on the analysis of oracle bone inscriptions, and there are no relevant public data sets. Although there are some traditional methods to try to recognize oracle bone inscriptions, such as the recognition method based on topological registration proposed by Gu Shaotong; Zhou Xinlun et al. proposed using graph theory and The method of identifying oracle bone glyphs based on stroke characteristics; Li Feng et al. proposed a method of using graph featu...

AI Technical Summary

Problems solved by technology

[0002] Oracle bone inscriptions are the earliest mature writing system discovered in China so far. They are the source of Chinese characters and the root of Chinese excellent traditional culture. However, the interpretation of oracle bone inscriptions is still difficult. There are few studies on the analysis of oracle bone inscriptions, and there are no relevant public data sets. Although there are some traditional methods to try to recognize oracle bone inscriptions, such as the recognition method based on topological registration proposed by Gu Shaotong; Zhou Xinlun et al. proposed using graph theory and The method of identifying oracle bone glyphs based on stroke characteristics; Li Feng et al. proposed a method of using graph features to identify oracle bone glyphs; Li Qingsheng et al. proposed a method of using graph isomorphism to identify oracle bone glyphs, but they often only use oracle bone glyphs as a whole It does not consider the similarity and internal structure between different characters, nor does it consider the problem of variant characters in oracle bone inscriptions that are different but still have the same glyph. However, there are a large number of variant characters in oracle bone inscriptions, so the practicability is limited

Images