home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!charon.amdahl.com!pacbell.com!decwrl!spool.mu.edu!agate!stanford.edu!morrow.stanford.edu!nova2.stanford.edu!mou
- From: mou@nova2.stanford.edu (Alex Mou)
- Newsgroups: comp.compression.research
- Subject: Re: Question about Quantization in the DCT domain
- Date: 11 Nov 1992 05:41:30 GMT
- Organization: Stanford University
- Lines: 31
- Message-ID: <1dq6eaINN892@morrow.stanford.edu>
- References: <1992Nov6.151148.7943@news2.cis.umn.edu>
- NNTP-Posting-Host: nova2.stanford.edu
-
- In article <1992Nov6.151148.7943@news2.cis.umn.edu> bedros@loon.cs.umn.edu (Saad Bedros 625 7542) writes:
- >
- >working in quantization for the DCT coefficients, I know I can apply
- >MAX quantizer for the ac coefficients with a laplacian source,
- >But for the DC term, a uniform quantization is best.
- >What is the best approach for quantizing the DC term:
- > 1- calculate the min and max, then quantizing using 8 bits after
- > normalization. in this case, the min and max should be transmitted.
- > 2- find a global max for the DC, normalize then quantize. I know in JPEG
- > they use 16 as a quantization level.
-
- From a hardware implementation point of view, finding a global max may
- need quite an amount of memories to accomodate all transformed blocks.
- This is the drawback I see in the second method.
-
- >
- >What is the advantage to set the range between -128,128 instead 0,255
- >before taking the DCT ? ( I think that is what they do in JPEG)
- >
-
- The advantage is to simplify H/W implementation.
- Hardware implementation usually uses 2's complement representation of
- numbers. That is the cause.
-
- >By the way, I was looking for a DCT in matlab, and I got one, not fast
- >but not slow if somebody is intersted.
- >
- >Saad Bedros
- >U of Minnesota
-
-
-