Region of Interest Coding for Aerial Video Sequences Using Landscape Models

Abstract

Depending on the use-case, though, it often is helpful to treat different areas of the image with different coding parameter sets or techniques even. In applications with focus on moving objects for example, a better resolution in the identified Regions of Interest (ROI) might help subsequent processing steps within a larger system. Existing video coding standards, such as MPEG-1,2,4 video or the ITU-T H.26x standards, only provide basic support for ROI coding. In e. g. MPEG-4 Video Object Planes (VOPs) and the separate encoding of these planes is included [1]. Unfortunately these features aren’t used to the extent possible, even though several real-life applications could be enhanced by such systems. Surveillance and videoconferencing tasks for example can benefit from a special ROI coding approach, wherein objects are automatically selected by e. g. motion (surveillance), color [2] (videoconferencing), shape or have been selected manually beforehand. Those regions are then coded with a higher quality than the rest of the picture. Especially for narrow-band transmission channels as used e. g. in aerial surveillance, it is important to keep the amount of data to be transmitted for the conduct of the task at hand to a minimum. In ROI coding it is one possibility to reduce this amount of data by degrading the quality of the parts of the image that are not as useful to the application.