Mining movies for song sequences with video based music genre identification system

This paper proposes a system that automatically locates and extracts songs from digitized movies and classifies the extracted songs according to their genre using a video-based genre identification algorithm. A song grammar particularly applicable to Bollywood movies has been proposed and used subsequently to construct a probabilistic timed automaton that differentiates songs from rest of the scenes in a movie. The proposed system has been implemented and test results indicate both high precision and recall in songs extraction and genre classification. Songs being a major driver in the success of south Asian movies, a potentially significant application of the proposed system lies in automatically mining the vast Bollywood movie archives. Furthermore, the genre identification part may also be used as an independent application in content-based music retrieval from online music archives.