Panel by Panel: Deep Learning-Based Comics Panel Extraction Using Mask Scoring

This research applies mask-scoring deep learning models to propose an original approach that can be used to recognize and index comic panels. The main objective is to locate and catalog each panel in a comic accurately to enhance the digital archive of comic books for search as well as retrieval. In this paper, we employing the mask scoring deep learning architecture, the panel recognition accuracy of conventional methods was significantly improved. Our statistics from tests resulted in 95% accuracy which is 10% higher than other known techniques. The results proved the applicability of the mask scoring models to alleviate the specific challenges related to the recognition of panel contents in comics, thus offering considerable benefits to digital preservation as well as enhancing the efficiency of relevant information search.