Using a 3D Shape Model for Video Coding

In this work, we present a model-aided coder that incorporates 3-D shape and motion information to improve the performance of a video coding system. The 3-D shape model of a rigid body is estimated from the video sequence directly. Using the shape model, the 3-D motion of the rigid body and the background global motion are also estimated for video coding. Experimental results show that bit-rate savings of 30%-40% are achieved at equal PSNR compared to the H.26L test model TML8.0. This corresponds to 2-3 dB improvements in PSNR when encoding at the same bit-rate.