Automatic music composition suitable for visual media

Our aim is to compose music that is suitable for expressing the emotions extracted from a visual medium such as a picture storybook. The emotional values obtained from pictures were inferred from the hue and color brightness of the pictures and those elicited from the story were calculated by matching between the words and a sentiment classification dictionary. Music was automatically composed using genetic algorithms by means of a suitability function based on the extracted emotion values. First, a single bar (or measure) of music was composed. Then, by using that bar of music as a “gene”, eight bars were created to form a complete period, or musical unit. Subjects evaluated the composed music in terms of “cheerful” and “gloomy” images, and the results obtained confirmed the validity of the suitability function.