N17040, WD 1 of Compact Descriptors for Video Analysis

The current media and entertainment domain is characterized by increasing volumes of content, large number of delivery channels and an ever-growing need for the relevant content to be accessed on demand or published quickly. This is very difficult without appropriate tools to manage content items, including search for object instances in video, categorisation of scenes and content grouping. Thus, the ability to generate and exchange compact and standardized descriptors in an interoperable and efficient way is considered a key enabler in this domain. An example application is scalable instance search to find a specific object instance in a very large video database (e.g. like a broadcaster might have). The industry needs video descriptors that enable performing this task with smaller descriptor size and shorter runtimes as compared to using CDVS (ISO/IEC 15938-13) in the video domain. Rigid objects (e.g., textured buildings) are of highest relevance, however object instances with more challenging properties (deformable, reflective and inherently varying spatial views) are also of interest.