The pel-recursive approach to motion estimation has been widely studied for compensating progressively scanned, moderate-resolution video. Although pel-recursive algorithms may not be suitable for application to interlaced high-definition television (HDTV), the underlying principle of backward motion compensation, upon which pel-recursive algorithms are based, can be exploited to improve the existing motion compensation algorithms. This paper proposes applying a backward approach to motion compensation to improve the performance of standard block-based algorithms for motion-compensated interlaced HDTV sequences. First, we describe a framework for motion compensation in which motion information is parameterized by a motion operator and a domain for that operator. Within this framework, we characterize the type of motion information represented by forward (e.g. block-based) and backward approaches to motion compensation. We propose a method for combining these two sources of motion information to form an optimal motion-compensated prediction. Simulations on two interlaced HDTV sequences demonstrate performance improvements between 1 and 2 dB over standard block-based methods.
Read full abstract