Teaching Neural Networks to Imitate Human Habits for Recognizing Anime Characters
Z. Y. Wang, Xin Sheng Mao, Zhaohui Zhang, Jiguang Zhang, Shibiao Xu, Xiaopeng Zhang · 2023
Recognizing anime characters poses a challenging task that differs from natural image recognition. Unlike natural images, anime images are generated by coloring and rendering existing sketches. Geometric features such as line shapes of anime images are more important for the task of anime character recognition, and they are the basis and keys to characterize the target. However, existing methods only focus on pixel and texture information, neglecting the importance of the anime characters' geometric features. In this work, we collect a dataset Anime-Sketch containing anime characters and corresponding sketches, and propose a multiscale residual dual-branch mutual learning fusion network for anime character recognition, which achieves consistency in representing the geometric shapes and color textures of anime images through multi-feature fusion, leading to significantly improved accuracy in recognizing anime characters. Extensive experiments on multiple datasets demonstrate the superior performance of our method.