Collaborative Video Annotation Based on Ontological Themes, Temporal Duration and Pointing Regions
Shah Khusro, Mumtaz Khan, Irfan Ullah · 2016
The development of standards like MPEG-7, MPEG-21 and ID3 tags in MP3 have recognized the importance of adding descriptions to multimedia content for the purpose of better organization and retrieval. However, these standards are limited in scope and are only suitable for closed world multimedia content where a lot of effort is put in the production stage. On the contrary, video content on the Web is of arbitrary nature, captured and uploaded in a variety of formats, with the primary aim of quick and easy sharing. The advent of Web 2.0 has resulted in the wide availability of different video-sharing applications like YouTube and has made video as a major content on the Web. These web applications not only allow users to browse and search multimedia content but also add comments and annotations which provide an opportunity to harvest wisdom of the crowd. However, these annotations have not been exploited to their fullest potential for the purpose of searching and retrieval. Video searching, ranking and recommendations could become more efficient if these annotations are made machine-processable under the guidance of domain-level ontologies. Moreover, associating annotations with a specific region, temporal duration and/or a specific theme of a video results in faster retrieval of required video scene of clip. In this paper, we propose a collaborative video annotation system that is based on temporal duration and pointing regions inside a video and also utilizing ontological themes of the selected domain. For the proof-of-concept development and evaluation, a comprehensive sports ontology (Cricket in this case) has been designed. The proposed system performs well in the context of free-text and ontological annotations. It performs at a higher level when browsing and searching related themes, scenes and objects as well as summarizing related themes, scenes and objects.