Object-Based Motion Parameterization Using Shape Adaptive Bilinear Warping

Abstract

The estimation of parameters describing accurately the motion of arbitrarily shaped image areas is the basis of the next generation of video coding standards and multimedia techniques. In this paper we describe techniques for object segmentation and use this object information to carry out a generalized scheme for shape adaptive block-matching in the context of content based coding. Object segmentation is performed by applying different approaches tailored to the addressed application. These approaches range from very-low complexity techniques for real-time segmentation to object segmentation in very complex images. Once masks of objects in the scene are available each foreground object is considered independently for motion estimation. A regular wire-grid is defined on the whole image and the wire-grid nodes are classified according to their position with respect to the contour of the considered object. The mesh is deformed from frame to frame by applying a bilinear warping while the motion parameters are chosen to minimize the local block-distortion. The presented technique has been evaluated in the context of object-based video coding.