Stroke Extraction Method for Calligraphy Characters Based on Enhanced Pix2pix Network

Extracting strokes from calligraphy characters is a key challenge in preserving traditional Chinese calligraphy digitally. In this paper, we propose a new method of stroke extraction using the Pix2pix network, which includes advanced attention mechanisms and a multi-scale discriminator. We add contextual transformer block and multi-head self-attention modules to the generator to better capture detailed strokes, focusing on edges and intersecting areas. Additionally, a multi-scale discriminator evaluates features at different resolutions, improving precision by distinguishing real images from generated ones. This method effectively solves the challenges of stroke extraction in calligraphy characters. The experimental results clearly demonstrate the effectiveness of our proposed method.