Model-based global motion estimation and compensation for very low bit-rate video coding in dynamic monocular systems

A segmentation-based hybrid video codec is introduced, that is able to work with sequences acquired by a moving camera. The global movement of the camera is compensated, allowing a working procedure in the codec that is closer to the case of a static background. The segmentation scheme consists of a relaxation algorithm followed by a bottom-up merging procedure. The region contour description is dramatically reduced through forward and backward region based motion estimation and compensation, applied over an image where the camera motion has already been compensated. Structural image formation is used to carry out a selective displacement field difference (DFD) coding. Although the encoded bit-rate is highly reduced, the reconstruction of image objects is improved, avoiding the characteristic blocking effects.