home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!hela.iti.org!usc!news.service.uci.edu!nntpsrv
- From: rohit@sloth.eng.uci.edu (Rohit Agarwal)
- Subject: ( Shimmering ) artifacts in wavelet based video compression ?
- Nntp-Posting-Host: sloth.eng.uci.edu
- Message-ID: <2B0844CE.8970@news.service.uci.edu>
- Newsgroups: comp.compression.research
- Organization: University of California, Irvine
- Keywords: Wavelet Transform, Video Compression
- Lines: 28
- Date: 17 Nov 92 01:03:10 GMT
-
- I have been doing some work with the compression of video by taking
- a wavelet transform, scalar quantization and entropy coding. I have implemented
- a video compression scheme that uses purely spatial techniques. No motion info
- is used. I applied my technique to the compression of several of the CCIR601
- MPEG test sequences and observed the following:
-
- In some of the sequences, particularly Tennis and Flower Garden, there
- was a noticable shimmering in the background. Viewed individually, each of the
- frames looked very good. For Tennis and Flower Garden, PSNR was about 35 dB on
- the average. However viewed as video, ther was a noticable shimmering artifact.
- Its as if the errors in each frame are highly uncorrelated with errors in other
- frames or maybe it has something to do with interlacing. It might also have
- been due to not using temporal information. I have seen some of the MPEG 2
- results from Kurihama and I noticed that there were a few techniques based on
- wavelets which did not seem to have these artifacts particularly Aware1 and
- Aware2.
- Can anyone give me some pointers as to the possible sources of this artifact.
- Interestingly enough, my technique did much better on Mobile and Calendar --
- individual frames had a lower PSNR than the other test sequences (about 30 dB),
- and there were noticable artifacts in the individual images viewed still (the
- white background of the calendar looked dirty gray) but viewed as a sequence
- it looked "pretty good" even at 1H.
-
- Also, the compression ratio was only 20:1 translating to a bitstream of about
- 8Mbps.
-
- rohit agarwal
-
-