Boundary-Control Vector (BCV) Motion Field Representation and Estimation By Using A Markov Random Field Model

A new motion field representation based on the boundary-control vector (BCV) scheme for video coding is examined in this work. With this scheme, the motion field is characterized by a set of control vectors and boundary functions. The control vectors are associated with the center points of blocks to control the overall motion behavior. We use the boundary functions to specify the continuity of the motion field across adjacent blocks. For BCV-based motion field estimation, an optimization framework based on the Markov random field model and maximum a posterior (MAP) criterion is used. The new scheme effectively represents complex motions such as translation, rotation, zooming and deformation and does not require complex scene analysis. Compared with MPEG of similar decoded SNR (signal-to-noise ratio) quality, 15-65% bit rate saving can be achieved in the proposed scheme with a more pleasant visual quality.