Affective Relationship between Color and Text in Arabic Comic Books

Abstract

Comic books are considered a heritage in many countries. The colorful depiction of annotated events has gained an increasing amount of interest over the past decade as the digitization process took over printed media, in addition to the abundance and variety of available data. Several applications have been devised in the field of computer vision and natural language processing to handle comic book pages. However, in this paper, we focus on the connection between the two, specifically, we compare the emotions that color and text separately imply. The theory of color and its effect on one's emotional state can be dated back to the 1800's. The language used in speech balloons is also written to a way to capture the audience's attention and manipulate their emotions. Throughout this paper, the color theory is applied to analyze the pages' emotional implication, and compared to the output of minSVM, a modified SVM classifier that accommodates imbalanced datasets, and a regular SVM, that are trained and implemented on the extracted text of a homegrown database to identify the emotions they convey. Using minSVM, we obtained a 91.26 % accuracy as opposed to an 89.66 % with SVM.