Anime characters have unique and exaggerated body structures, which pose a challenge for artists to accurately draw their human anatomy. In this paper, we propose a method for human body structure extraction of anime characters based on deep learning and discover cross-domain relations with generative adversarial networks (DiscoGAN)[1]. Our approach utilizes Generative Adversarial Networks (GANs)[2] to automatically extract the human body structure from anime character images. To train our model, we collected a dataset of anime character images and corresponding human body structure. We also created our own dataset of human body structures using Maya[3] software. The dataset consisted of geometric body blocks that were animated and rendered using frame-by-frame rendering techniques, resulting in images of the body blocks in various poses. Although our current dataset only contains 1000 images, we plan to expand it in future research. The proposed method can accurately extract the human body structure from anime character images, which can be used as a reference for beginners to draw more anatomically accurate human figures.

The popularity of social networking services and smartphones has led to a growing demand for more advanced photo processing techniques. In the field of image generation, Generative Adversarial Networks (GANs)[2] have shown remarkable results. Our proposed method is based on deep learning, which can learn the complex patterns of human body structure and automatically extract it from anime character images. This research is expected to contribute to the field of drawing by providing a tool that can significantly reduce the learning period or provide a new study method for drawing beginners.

Our proposed method offers a new perspective for extracting human body structure from anime character images, which can be useful for beginners. The effectiveness of our proposed method is evaluated, demonstrating its potential to accurately extract human body structures from anime character images.