Analyzing Emotional Impact of Speech Balloons on Comic Characters: A Study of Arousal Levels

The dialogue between the protagonists in comics, where much of the information is communicated, usually comprises speech balloons. Thus, identifying the speech balloons and connecting them to the speaker is a crucial step. The speech balloon's shape also reveals information about the speaker's emotions. The main aim of this study is to recognize the emotional arousal of the comic character with the shape of the speech balloon. To achieve this new model has been designed with a Convolutional Neural Network and SVM, which helps to detect the features and categorize them into an emotional state associated with the shape of the speech balloon. With the proposed approach the authors have achieved 92% accuracy which is comparable accuracy with the previous work. As a result of the feature selection, it was determined that the final characteristics were more useful for estimating the shape of the speech balloon. From the experimental results, it can be observed that a higher precision rate is achieved for the surprising class i.e. 92.60%, and a higher recall rate is achieved for the anger emotion class i.e. 93.10 and a higher Fl score is achieved for the polite emotion class i.e. 93%.