Extraction of Eyes and Mouth for Anime Characters using Color and Position Information

Abstract

In this paper, we propose facial parts extracting method for the eyes and mouth of an animation character by focusing on illustration characteristics. Typically, machine learning has been applied to the extraction of facial parts; however, the learning cost, such as training time and creating datasets, is an issue in this approach. In contrast, our method identifies and extracts facial parts based on heuristics obtained from illustration characteristics, in particular color and position information. We also demonstrate that our method can be used to automatically create a dataset for training a machine-learning-based extraction method and is thus expected to reduce the burden of manual dataset creation. In this paper, we evaluated our method through experiments using illustrations available on the web. Results showed that the detection accuracy of some facial parts is high, but our method is still in a developing stage to be used as automatically creating a dataset.