Two-Dimensional Shape Coding

Abstract

content-based storage and retrieval have to allow access to video data based on object descriptions, where objects are described by texture, shape, and motion. Studio and television postproduction applications require editing of video content with objects represented by texture and shape. For collaborative scene visualization like augmented reality, we need to place video objects into the scene. Mobile multimedia applications require content-based interactivity and content-based scalability in order to allocate limited bit rate or limited terminal resources to fit the individual needs. Security applications benefit from content-based scalability as well. All these applications share one common requirement: video content has to be easily accessible on an object basis. MPEG-4 Visual enables this functionality. The main part of this chapter describes MPEG-4 shape coding, the content-based interactivity enabling tool.Given the application requirements, video objects have to be described not only by texture, but also by shape. The importance of shape for video objects has been realized early on by the broadcasting and movie industries employing the so-called chroma-keying technique, which uses a predefined color in the video signal to define the background. Coding algorithms like objectbased analysis-synthesis coding (OBASC) [30| use shape as a parameter in addition to texture and motion for describing moving video objects. Second-generation image coding segments an image into regions and describes each region by texture and shape [28]. The purpose of using shape was to achieve better subjective picture quality, increased coding efficiency as well as an object-based video representation. Two types of VOs are distinguished. For opaque objects, binary shape information is transmitted. Transparent objects are described by gray-scale a-maps defining the outline as well as the transparency variation of an object. MPEG-4 Visual is the first international standard allowing the transmission of arbitrarily shaped video objects (VO) [21 J. Each frame of a VO is called video object plane (VOP