2001 IEEE International Workshop on Multimedia Signal Processing (MMSP-01) 
October 3-5, 2001, Cannes, France

"ADAPTIVE PICTURE SLICING FOR DISTORTION-BASED CLASSIFICATION 
OF VIDEO PACKETS"

Authors: Enrico Masala, Davide Quaglia, Juan Carlos De Martin

The video sequence was encoded using TM-5 Encoder. Then the MPEG-2 bitstream was divided into packets. For the test cases "random marking" and "adaptive distorsion-based marking" a portion of packets was marked as "premium": in the "adaptive distorsion-based marking" we marked those packets that, if lost, would be recovered hardly, given a concealment technique in the decoder.

Then a loss pattern was applied to the not-marked packets in all test cases (of course in "no marking" all packets are not-marked). Damaged bitstreams were decoded by a modified version of the ISO Reference Decoder in order to implement a simple concealment technique: it replaces the lost portion of a picture with the area taken from the previous picture at the same position.

PSNR was computed on decoded-concealed video data from damaged bitstreams with respect to decoded video from the loss-free bitstream.

Such decoded-concealed video data were re-encoded in MPEG-1 to make these demo files: in fact the most famous media players (both in Windows and in Linux) does not read MPEG-2.

Click on numbers to open/download demo files (about 1 MB for each file):

loss-free bitstream

Premium share
No marking
random marking
adaptive distorsion-based marking
20%
33.87
35.64
44.02
PSNR values (in dB) for packet-loss rate=7%

 
Premium share
No marking
random marking
adaptive distorsion-based marking
20%
32.42
33.32
37.11
PSNR values (in dB) for packet-loss rate=15%