On the use of computable features for film classification

Abstract

This work presents a framework for the classification of feature films into genres, based only on computable visual cues. We view the work as a step toward high-level semantic film interpretation, currently using low-level video features and knowledge of ubiquitous cinematic practices. Our current domain of study is the movie preview, commercial advertisements primarily created to attract audiences. A preview often emphasizes the theme of a film and hence provides suitable information for classification. In our approach, we classify movies into four broad categories: Comedies, Action, Dramas, or Horror films. Inspired by cinematic principles, four computable video features (average shot length, color variance, motion content and lighting key) are combined in a framework to provide a mapping to these four high-level semantic classes. Mean shift classification is used to discover the structure between the computed features and each film genre. We have conducted extensive experiments on over a hundred film previews and notably demonstrate that low-level visual features (without the use of audio or text cues) may be utilized for movie classification. Our approach can also be broadened for many potential applications including scene understanding, the building and updating of video databases with minimal human intervention, browsing, and retrieval of videos on the Internet (video-on-demand) and video libraries.