Three-dimensional compression of video sequences

The two-dimensional, still image compression standard JPEG was extended to three dimensions and used to compress test standard video sequences. The 3-D video compression algorithm was used to compress the black-and-white, digital video sequences 'football' and 'Claire'. For the SIF video sequence 'Claire', a compression ratio of 200 was achieved at a Peak Signal to Noise Ratio (PSNR) of 29 dB, and a compression ratio of 100 was achieved at a PSNR of 35 dB. Ohm has reported 200 times compression ratio for 3-D video compression for similar video sequences. Ohm's PSNR was 27 dB, which is 2 dB more distortion than that achieved in this research.; Rate-distortion performance for 'Claire' and for the interlaced, NTSC video sequence 'football' were determined. Reconstructed pictures for each sequence are included in this thesis.; The 3-D quantization matrix and the 3-D coefficient transmission order were motivated by the JPEG compression algorithm. The rate-distortion optimization and motion compensation algorithms extended MPEG-2 methods to three dimensions.; The 3-D coefficient transmission order strongly effected the rate-distortion performance. Ordering by the increased probability of a coefficient equalling zero is the motive for JPEG's transmission order, and produced the best results. But the overhead required to transmit the transmission order strongly reduced the compression ratio. A nearly optimal transmission ordering was determined based on the root sum square of the temporal frequency and the horizontal spatial frequency. The rate-distortion performance of the nearly optimal transmission order was within 10% of the optimal order's performance. Ordering by spatio-temporal frequency, in the manner of JPEG, produced compression ratios approximately one-half of the nearly optimal transmission order. Poor energy compaction for the vertical spatial frequency coefficients caused the suboptimal performance when coefficient transmission order was based on increasing spatio-temporal frequency.; There were two steps in the 3-D video compression algorithm that were not associated with MPEG or JPEG algorithms. First, the transform used was an adaptive 3-D Approximate Karhunen-Loeve Transform, that extended the still image compression work of Reed and Lan to three dimensions. The adaptive 3-D AKLT reduced distortion 5% at the same rate as that achieved by the 3-D DCT. Secondly, areas of the video sequence that had a large amount of temporal activity were segregated out for special rate-distortion optimization. As a result, visible distortion to the blinking eye portion of the 'Claire' sequence was removed.