Attention-Based Feature Fusion Network for Comic Character Personality Prediction

As a popular form of entertainment and art, comics provide readers with rich stories and visual enjoyment. The types of characters in comics are rich and diverse with unique charm. The appearance, lines and actions of comic characters are all part of the role shaping. In order to better analyze comic characters' personalities through comic images and lines, and further obtain information between comic text and image data, this paper proposes an attention-based fusion neural network, which consists of three parts: feature extraction, cross-modal feature learning, and personality prediction. The effectiveness of this network is validated through experiments based on the Manga109 and Manga109Dialog datasets.