Compression of cubic-panorama datasets with spatially consistent representation

Abstract

Efficient compression of cubic-panorama datasets is crucial to reducing the required storage space and transmission bandwidth. In traditional coding schemes, cubic-panorama datasets are treated as planar image sequences with a rectangular support. In this way, it is inevitable to sacrifice some coding efficiency because of the inconsistency on the boundaries of connected side images. We have developed a spatially consistent representation for cubic panoramas, and applied a motion-compensated temporal filtering (MCTF) coding scheme to compress cubic-panorama datasets. Specific approaches for constructing the reference blocks on the corners of cubes are designed for motion estimation. The search for motion vectors in reference frames can be naturally extended across side-image boundaries into neighbor side images. The spatially consistent representation of cubic panoramas eliminates image boundary constraints for motion vector search. The search for motion vectors as well as matching reference blocks can be extended well beyond the side image boundaries in all four directions. Better matched reference blocks can be obtained to further reduce the prediction errors and improve the compression efficiency. The compression scheme is adapted to the features of cubic-panorama datasets. The experimental results of applying the proposed compression scheme to coding sample cubic-panorama datasets are presented. It is shown that superior coding performance is achieved with the spatially consistent representation compared with the generic representation of cubic-panorama datasets