Qin Seal Script Character Recognition with Fuzzy and Incomplete Information

Abstract

The dependable and efficient identification of Qin seal script characters is pivotal in the discovery, preservation, and inheritance of the distinctive cultural values embodied by these artifacts. This paper uses image histograms of oriented gradients (HOG) features and an SVM model to discuss a character recognition model for identifying partial and blurred Qin seal script characters. The model achieves accurate recognition on a small, imbalanced dataset. Firstly, a dataset of Qin seal script image samples is established, and Gaussian filtering is employed to remove image noise. Subsequently, the gamma transformation algorithm adjusts the image brightness and enhances the contrast between font structures and image backgrounds. After a series of preprocessing operations, the oriented gradient histograms (HOG) features are extracted from the images. During model training, different weights are assigned to classes with varying sample quantities to address the issue of class imbalance and improve the model's classification accuracy. Results show that the model achieves an accuracy of 95.30%. This research can help historians quickly identify and extract the text content on newly discovered Qin slip cultural relics, shortening the cycle of building a historical database.