Proceedings Article | 14 March 2013
KEYWORDS: Video, Semantic video, Video coding, Video processing, Image segmentation, Image compression, Visualization, Photography, Cameras, Computer programming
In recent years, in order to reduce semantic gap which exists between high-level semantics and low-level features of
video when the human understanding image or video, people mostly try the method of video annotation where in signal’s
downstream, namely further (again) attach labels to the content in video-database. Few people focus on the idea that: Use
limited interaction and the means of comprehensive segmentation (including optical technologies) from the front-end of
collection of video information (i.e. video camera), with video semantics analysis technology and corresponding
concepts sets (i.e. ontology) which belong in a certain domain, as well as story shooting script and the task description of
scene shooting etc; Apply different-level semantic descriptions to enrich the attributes of video object and the attributes
of image region, then forms a new video model which is based on Video Object Plan (VOP) Coding. This model has
potential intellectualized features, and carries a large amount of metadata, and embedded intermediate-level semantic
concept into every object. This paper focuses on the latter, and presents a framework of a new video model. At present,
this new video model is temporarily named “Video Model of Semantic-Preloaded or Semantic-Preload Video Model
(simplified into VMoSP or SPVM)”. This model mainly researches how to add labeling to video objects and image
regions in real time, here video object and image region are usually used intermediate semantic labeling, and this work is
placed on signal’s upstream (i.e. video capture production stage). Because of the research needs, this paper also tries to
analyses the hierarchic structure of video, and divides the hierarchic structure into nine hierarchy semantic levels, of
course, this nine hierarchy only involved in video production process. In addition, the paper also point out that here
semantic level tagging work (i.e. semantic preloading) only refers to the four middle-level semantic. All in all, this
research was unfolded is based on analyzed the characteristic of the existing video mode, and with reference to MPEG
series standard.
