Interframe coding using deformable triangles of variable size

Motion compensation using deformable triangle patches has been successfully used for low bit rate coding of videophone sequences. They were also shown to be particularly efficient for interframe coding of MRI sequences, for which the difference between image slices can be well modeled by affine deformations. Regular triangular meshes were used in previous work. In this paper we present a quadtree decomposition algorithm to generate a triangle mesh for which smaller triangles are used in image areas where the motion or deformation is more complex. The decomposition algorithm uses a strategy based on rate-distortion theory. The cost of using a new grid point, i.e., the bit rate, is compared to the reduction in prediction error energy that would result if that grid point is used. Grid points are recursively added to areas where the reduction in prediction error is more significant. Results show that using variable size triangular patches increases the SNR of the motion-compensated image by 0.5 to 0.75 dB while reducing the number of grid points by 15 to 25% when compared to a regular triangular mesh