A Video Scene Detection Method Using Audio Information

Abstract

The role of audio in the context of multimedia applications involving video is becoming increasingly important. In this paper, an approach to automatic segmentation and classification of audiovisual data based on audio content analysis is proposed. Specifically, an audio classification scheme is developed to partition the sound-track of a video into homogeneous audio segments such as speech, music and speech with music background signal. Audio features which are extracted from both time and frequency domain are employed to ensure the feasibility in scene change detection. A hierarchical classification approach is applied for fast segmentation and detection. A Support Vector Machine (SVM) is firstly used to detect scene with music. Then Gaussian Mixture Model (GMM) is adopted to classify the rest scenes into either scene containing speech only or scene consisting of speech with music background. The experiments on real documentary videos show that proposed approach provides satisfactory detection rates.