Automated Classification of Comics into Genres using CNN-SVM Model: A Study on Visual Storytelling

Comics utilize a combination of sequential panels or frames to narrate a story through visual art. With the use of text, dialogue, images, and sound effects, comics can convey a wide range of narratives across various genres, such as science fiction, romance, horror, comedy, and superheroes etc. The purpose of this study was to utilize the CNN-SVM model to classify comic pages into various genres. The results demonstrate that the model can effectively and accurately classify comic pages into their respective genres, achieving an overall accuracy of 93.26%. The precision, recall, and F1 score for each genre are all above 90%, indicating the model performs well across all genres, with Battle having the highest F1 score of 94.50% and Romantic Comedy having the lowest F1 score of 92.72%. Nonetheless, all genres have an F1 score above 90%, indicating the model's effectiveness in recognizing each genre. The micro-average, macro-average, and weighted average for all genres are nearly identical, suggesting the model's consistency across all genres. Overall, these results demonstrate that the CNN-SVM model can be useful in automating the classification process of comic pages into genres, which can have practical applications such as content recommendation and analysis.