Frame Selection for Automatic Comic Generation from Game Log

Recently, we have presented a comic generating system that visualizes an online-game play. Our system was inspired by a former work of Shamir et al. However, comics generated in their work can have series of similar frames when multiple actions occur near each other in both time and space. In this paper, we first present a frame-selection module that uses Habituated Self-Organizing Map. Our method prevents comic readers from boredom by getting rid of resemble consecutive frame candidates. We then evaluate the method by a subject experiment using a play from the ICE, an online-game developed in our laboratory. Experimental result confirms that our method is effective in making output comics more interesting than a baseline method.