Automatic actors detection in musicals

摘要

Bollywood movies have become an important part of South Asian culture. Video musical pieces are integral parts of these movies. The growing ease of hosting multimedia contents online is resulting in a huge number of movie video songs available online. A common user usually likes to search a particular music piece, based on different criteria e.g. genre of a video song, or on-screen actor performing on a video song. Such information, in current implementation, is attached manually as a textual caption with the video songs. This textual caption with a video song is highly unreliable, especially when a video song is to be uploaded by an ordinary user, which is usually the case with online video songs. In this paper, we present an automatic approach of detecting the actors vocalizing a video song, who are usually the stars of the movie. Our approach is based on detecting actors faces from different parts of a video song and clustering the faces based on faces similarity. We tested the proposed actors detector on 30 video songs, belonging to different genres. An overall accuracy of 83.34% and 80% has been achieved on first and second main actors, respectively.