State-of-the-art and future challenges in video scene detection: a survey

摘要

In the last 15 years much effort has been made in the field of segmentation of videos into scenes. We give a comprehensive overview of the published approaches and classify them into seven groups based on three basic classes of low-level features used for the segmentation process: (1) visual-based, (2) audio-based, (3) text-based, (4) audio-visual-based, (5) visual-textual-based, (6) audio-textual-based and (7) hybrid approaches. We try to make video scene detection approaches better assessable and comparable by making a categorization of the evaluation strategies used. This includes size and type of the dataset used as well as the evaluation metrics. Furthermore, in order to let the reader make use of the survey, we list eight possible application scenarios, including an own section for interactive video scene segmentation, and identify those algorithms that can be applied to them. At the end, current challenges for scene segmentation algorithms are discussed. In the appendix the most important characteristics of the algorithms presented in this paper are summarized in table form.