Detection and classification of speech balloons in comic images

Abstract

Today, the demand is increasing for comic contents on cellular phones and speech software for the visually impaired. When speech software reads aloud a comic character's speech, it is useful for both the visually impaired and unimpaired to have the character's voice injected with his/her feeling, which is inferred from types of speech balloons. As a result, comic contents come to life. In this research, a method has been developed to detect speech balloons on comic pages and then classify them into four types. In this method, speech balloon candidates are extracted based on speech text information detected by AdaBoost, and then speech balloons are selected and classified using SVM. Experimental results show that the proposed method successfully detected and classified 86 percent of 2844 speech balloons.