The Study of Chinese Calligraphy Font Style Based on Edge-Guided Filter and Convolutional Neural Network
Leilei Xiang, Yue Zhao, Guojun Dai, Ruoyun Gou, Hua Zhang, Jiankai Shi · 2020
Chinese calligraphy is a unique visual art and one of the material bases of Chinese traditional cultural heritage. However, due to the passage of time, ancient calligraphy works have been weathering and damages, so it is necessary to utilize advanced technologies to protect those works. The image features are not only used to represent the image content or structure but also used to express the style of images. In our work, we propose an extraction method to obtain traditional Chinese calligraphy (TCC) features of five major styles by fuzzy threshold segmentation and edge-guided filter, and then discriminate TCC styles using a convolutional neural networks (CNNs) of deep learning framework. At the same time, we use cross-validation to make the training results more convincing, and after considering the number of convolutional layers, training set, and training time, we chose VGGNet and ResNet as classifiers. Ultimately, we can achieve high-accuracy classify within less running time. And the method that we propose can also be applied to other image style classification problems.