Oracle Bone Script Similiar Character Screening Approach Based on Simsiam Contrastive Learning and Supervised Learning

Xinying Weng,Yifan Li,Shuaidong Hao,Jialiang Hou

2024-08-13

Abstract:This project proposes a new method that uses fuzzy comprehensive evaluation method to integrate ResNet-50 self-supervised and RepVGG supervised learning. The source image dataset HWOBC oracle is taken as input, the target image is selected, and finally the most similar image is output in turn without any manual intervention. The same feature encoding method is not used for images of different modalities. Before the model training, the image data is preprocessed, and the image is enhanced by random rotation processing, self-square graph equalization theory algorithm, and gamma transform, which effectively enhances the key feature learning. Finally, the fuzzy comprehensive evaluation method is used to combine the results of supervised training and unsupervised training, which can better solve the "most similar" problem that is difficult to quantify. At present, there are many unknown oracle-bone inscriptions waiting for us to crack. Contacting with the glyphs can provide new ideas for cracking.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the screening and identification of similar characters in oracle bone inscriptions. Specifically, due to the diverse forms and different styles of oracle bone inscriptions, traditional identification methods are difficult to identify similar character forms efficiently and accurately. For this reason, the paper proposes a new method, which combines supervised learning and self - supervised learning techniques to improve the recognition accuracy and efficiency of similar - shaped characters in oracle bone inscriptions.

### The main contributions of the paper include:
1. **Fusion of supervised learning and self - supervised learning**: The paper proposes a comprehensive model based on ResNet - 50 self - supervised learning and RepVGG supervised learning. The results of the two learning methods are integrated through the fuzzy comprehensive evaluation method, so as to better solve the difficult - to - quantify task of "the most similar".
2. **Image enhancement technology**: Before model training, the image data is pre - processed, including random rotation, self - square - map equalization theoretical algorithm and gamma transformation, which effectively enhances the learning of key features.
3. **Feature extraction and matching**: The deep convolutional neural network (such as RepVGG) is used to extract features from oracle bone inscription character data, and the feature vectors are compared and matched for similarity under the SimSiam framework.
4. **Fuzzy comprehensive evaluation**: The fuzzy comprehensive evaluation method is used to evaluate the results of supervised training and self - supervised training, which improves the credibility and accuracy of the results.

### Specific implementation steps:
1. **Data pre - processing**: Gray - scale processing, gamma transformation and histogram equalization are performed on the original image to enhance the key features of the image.
2. **Feature extraction**: The RepVGG model structure is used to extract features from oracle bone inscription character data.
3. **Similarity matching**: The extracted feature vectors are input into the SimSiam framework and matched by the cosine similarity function to find the most similar image.
4. **Model training**: Combining the supervised loss function and the self - supervised learning objective function, the training results of the two models are synthesized by the weighting method to obtain the best training effect.
5. **Fuzzy comprehensive evaluation**: The training effect of each model is evaluated by fuzzy comprehensive evaluation to further improve the reliability of the results.

Through these methods, the paper aims to provide an efficient and accurate similar - character screening tool for oracle bone inscription research, and provide new ideas and technical support for the research and interpretation of ancient characters.