Decoding:

The decoding process is shown Figure 4. Each block of data is treated independently by the client. The joint-angle parameters are applied to the articulated model. The resulting model projected to the camera of each depth stream, giving the proxy depth maps. The residue maps are recovered from the R frames of the packet. The residue map is added to the proxy depth map to get the decoded depth map for that view and frame. For D frames, the residue differences recovered from the packet are added to the current residue map $R_i$ to get the next residue map $R_{i+1}$, which is added to the proxy depth map for that frame to get the decoded depth values as given in equation 2. The foreground mask is needed to keep track of the changing silhouette of the actor. The depth map of frame $i+1$ may include pixels not in frame $i$. If frame $i+1$ is a D frame, the current average residue value is used as the reference for the residue difference. If the incremental frame arrives, the bit-plane for it is assembled and added to the current running residue $R_i'$. The improved quality results till the end of the current block.