WHOA!! Backup the JPG truck...

[P3D Jonathan Gross <catalyst@xxxxxxxxxxxx>]

Several members responded to my comments on Dan Shelley's cleverly
titled posting.  Lets see if I can clarify them in a single response.

> I agree with you, but what are the 
> advantages left from JPEG for image
> compression?

What does this question mean?? =)

There are two major advantages to JPEG over alternative formats:
1. It is a well accepted standard.
2. It allows variable compression while maintaining spatial resolution.

Because there are many well accepted defacto standard image formats, if
you don't use JPEG's image compression, I think you are giving up most
if the advantage.

> >If people published their 
> > JPEG and JPS images with 0% compression 
> > and 2X1024X768 resolution, then it might 
> > not be so bad.  But is that what people 
> > are doing?

> Come to my house, and see my personal 3D JPS images...
> they are quite nice, and so HUGE that until someone steps up to the
> plate to pay the bills for me, they will NOT be on my web site for
> people to download.

Then I guess the answer is no, people are not publishing high quality
images on the web in JPEG or JPS format, although they may be creating
them in the privacy of their own homes.  Perhaps this is why Dr. T and
others who believe what's on the web is the best their is, find the
electronic images so disappointing.

> Yes, that fine... But, this discussion was not about COMMERCIAL stereo
> images... It was about anaglyphs on the web... I don't understand the
> jump.

Im sorry that it seemed like a jump.  Its actually a series of related
steps that I did not make explicit:

As a backdrop, I thought that photo-3d tries to advance and promote the
art and science of stereoscopic imaging.  I though the discussion was
about the quality of electronic stereoscopic images on the web, and in
particular, anaglyphs in JPEG format.  Although I agree with you that
good quality anaglyphs can be posted in JPEG, you indicated that they
must have high spatial and chromatic resolution (i.e. be screen-size and
use 0% compression).  High quality images that would be acceptable to
the general public are commercial quality images as opposed to amateur
quality hobby images  (this does not mean that hobbyists cannot create
processional quality images).  I extended your techniques to stereo
pairs as an alternate class of stereoscopic images published on the
web.  I agreed that high resolution JPEG images with 0% compression
might yield good quality.  If  you dont compress the images with JPEG,
then why use it;  there are image compression techniques which will
provide compression without any loss of quality.  I believe that most
JPEG and JPS images on the net are poor quality.  Poor quality images on
the web do not help promote the advancement of electronic stereoscopic
imaging.  Perhaps the stereo community should look at some alternatives
which would allow high quality compressed stereoscopic images to be
published on the web.  Forcing authors to choose between poor quality
and unwieldy size does not advance the cause, IMO.


> >You'll get between 3:1 and 5:1 compression with no loss of
> >important boundary data.
> 
> Note, however, that the lossless compression ratio is dependent on the
> input data...

Yes, that is why I stated it as a range.

> Has the 3:1 to 5:1 range stated here proven out over a large
> and diverse set of images?

This has been my experience.  If there is a standard set of stereoscopic
images to which I can apply the method, I will try it and report the
results.

> I'm not at all sure why anyone would want to store separate 8 bit
> channels other than scientific curiosity. What's the point?

The point is that data compression techniques work by finding redundant
patterns in the data and substituting short codes to replace long
sequences of bits.  If you take an image with 24-bit color and separate
it into 8-bit channels, each of the channels will usually contain much
more redundancy the original 24-bit image.

Take a long list of full names.  If you try and compressing it by coding
all of the John Smiths as 1, and all of the Jane Does as 2  there is
likely to be little redundancy, depending on where the list of names
came from.

Now separate the list into a list of first names, and a list of last
names.  On the first name list, code all Johns as 1, Janes as 2, Bobs as
3 , and on the last name list, code all Smiths as 1, Does as 2, Berlins
as 3   These lists contain a lot of redundancy, and may be able to be
highly compressed.  When you want the full list back, expand the first
and last name lists, and put them back together. Of course, a computer
does it a lot easier and faster.

This is a scientific (or mathematical) curiosity, but it is also very
useful.

On the other hand, lossy compression techniques, such as JPEG and JPS,
when they dont find enough redundancy in an image,  blur the colors to
create some.  Apply the name list metaphor to JPEG, and Jonathan Gross
and Larry Berlin might come out John Grossman and Lance Berman.  Close
enough? 

The point is that I think low quality stereoscopic images on the web are
a turn-off for both seasoned stereophiles and the newly curious.

Jon Gross


------------------------------