Analysis of Partial Semantic Segmentation for Images of Four-Scene Comics

Ways of understanding human creations with the help of artificial intelligence (AI) have increased; however, those are still known as being one of the most difficult tasks. Our research challenge is to find ways to understand four-scene comics through AI. To achieve this aim, we used a novel dataset called “Four-scene Comics Story Dataset”, which is the first dataset made by researchers and comic artists to develop AI creations. In this paper, we focused on the partial semantic segmentation of features such as eyes, mouth, or speech balloons. The semantic segmentation task of comics has been difficult because of the lack of annotated comic dataset. To solve this problem, we utilized the features of our dataset and easily created annotated dataset. For the semantic segmentation method, we used a model called DeepLabv3+. The effectiveness of our experiment is confirmed by computer simulations showing the segmentation result of test images from four-scene comics.