Color-Based Content Retrieval of Animation Movies: A Study

Abstract

This paper discusses a methodology for the semantic characterization of the animation movie color content in the video indexing framework. Color is a major feature of animation movies: almost each movie uses a particular color palette which is related to its artistic content. To characterize the color content we use a fuzzy symbolic representation of several low-level statistical parameters which are extracted from the movie global color distribution (color signature). The derived semantic/symbolic information is regarding the movie's predominant hues, the Itten's color contrasts, the color harmony schemes and color relationship. The interest in these descriptions is twofold: first we provide a content analysis tool for the animation experts and second we provide the movie semantic content annotation which is required by the indexing task. The discriminating power of our features as prospective semantic indexes is tested using a k-means clustering. Moreover, to overcome the difficulty of comparing various types of data, also required by the content-based retrieval task, we have developed an efficient methodology for comparing movie content. It was inspired from the graphical representation of the color gamut used in printing devices. Several experimental tests were conducted on an animation movie database.