A New Comic Image Segmentation and Adaptive Differential Evolution Algorithm with Different Times Characteristics

Abstract

Comic scene segmentation is crucial in understanding and analyzing visual storytelling, as it involves identifying and separating distinct elements within a sequence of panels. This paper proposes a novel segmentation approach, Frog Leap Differential Time Series Segmentation (FLDTSS), tailored for analyzing comic images, which often contain complex visual storytelling elements such as expressive characters, dynamic speech bubbles, and background effects. By leveraging time-series features across sequential comic panels, FLDTSS integrates both spatial and temporal cues for more context-aware segmentation. The method was tested on a diverse set of cartoon panels and achieved a precision of 91.6%, recall of 88.3%, and an F1-score of 89.9%, outperforming traditional methods such as Otsu Thresholding (F1-score: 70.6%), Edge-based Canny (76.1%), K-means Clustering (77.8%), Watershed (80.6%), and even Genetic Algorithm-based segmentation (83.2%). The segmentation time for FLDTSS was 1.22 seconds, demonstrating computational efficiency compared to more intensive evolutionary methods. Simulation results showed the model's ability to extract meaningful narrative components such as characters, speech bubbles, emotional cues, and visual effects, with background occupying ~55% of the segmented area, character regions ~22%, and speech bubbles ~8%. This study confirms FLDTSS as a powerful and scalable technique for semantic segmentation and narrative interpretation in visual storytelling formats like comics.