Automatic Mash Up Music Video Generation System by Remixing Existing Video Content

Abstract

Music video is a short film which presents a visual representation of recent music. In these days, there is a trend that amateur users create music video in the video sharing website. Especially, the music video which is created by cutting and pasting existing video is called mashup music video. In this paper, we proposed the system that users can easily create mushup music video by using existing music videos. In addition, we conducted assessment evaluation experiment for our system. The system firstly extracts music features and video features from existing music videos. Then, the each feature is clustered and the relationship between each feature is learned by Hidden Markov Model. At last, the system cuts learned video scene which is the closest feature among learned videos and pastes it synchronizing with input song. Experiment shows that our method can generate more synchronized video than a previous method.