Deep Learning for Comic Book Emotion Analysis

Deep learning techniques have been applied successfully in a number of fields, including computer vision and image processing. This study presents a comprehensive approach to analyzing the sentiment conveyed in comic book panels through the emotions depicted on characters' faces. We hypothesize that it is possible to accurately interpret emotional content using deep learning techniques even without a pre-existing sentiment dataset. Using optical character recognition and pretrained sentiment analysis models, the basis of an NLP model was formed to comprehend emotional context from characters' dialogues and thoughts. A neural network then categorizes emotions exhibited by characters' facial expressions. Our findings confirm that sentiment analysis can indeed be performed on comic book data, with tests on the Digital Comic Museum dataset demonstrating sentiment analysis efficacy of 89%. Then, the optimal configuration of convolutional neural networks was identified with 7x7 filters, 200 neurons, and 64 filters per layer, achieving an accuracy of 86%. This research advances the capability of facial recognition technology, expanding its application from humans to fictional characters in comic books. It also sets the groundwork for future research to generate datasets when specialized data is not available, demonstrating the practicality of performing sentiment analysis on comic book faces.