Ron N. wrote:
> Richard Owlett wrote:
> > I understand that JPEG is a "lossy" compression method.
> >
> > For color images is loss in:
> >   1. resolution
> >   2. color depth
> >      a. would that include contrast
> >   3. resolution AND color depth
>
> Yes.  JPEG transforms the data to another vector basis where
> everything is affected.  The RGB to YUV conversion plus
> downsampling the UV throws away color resolution.  Then
> the DCT plus non-uniform quantization throws away both depth
> and resolution of both color and luminance in something like

It's worth noting that non-uniform does not mean non-linear.
Different channel/frequency/frequency combinations may
have different quantum sizes, but on decoding, the output
values are still proportional to the number of quanta.
That said, that doesn't mean that an encoder, told to be stingy
with storage space, might not round 2.7 to zero in some contexts
and round 1.6 to 2 in others.

Ron N. wrote:
> Richard Owlett wrote:
> 
>>I understand that JPEG is a "lossy" compression method.
>>
>>For color images is loss in:
>>  1. resolution
>>  2. color depth
>>     a. would that include contrast
>>  3. resolution AND color depth
> 
> 
> Yes.  JPEG transforms the data to another vector basis where
> everything is affected.  The RGB to YUV 

AHAH, YUV is key to all
didn't recognize, therefore Goggled

Oh, so that was what my father wanted to get across to me in 50's
[he was a Kodak engineer at _Camera Works_ ;]

thanks

 > conversion plus
> downsampling the UV throws away color resolution.  Then
> the DCT plus non-uniform quantization throws away both depth
> and resolution of both color and luminance in something like
> the frequency domain, given that some of the higher frequencies
> are sometimes quantized to zero bits, depending on the
> compression ratio.  etc.
> 
> 
> IMHO. YMMV.

Richard Owlett wrote:
> I understand that JPEG is a "lossy" compression method.
>
> For color images is loss in:
>   1. resolution
>   2. color depth
>      a. would that include contrast
>   3. resolution AND color depth

Yes.  JPEG transforms the data to another vector basis where
everything is affected.  The RGB to YUV conversion plus
downsampling the UV throws away color resolution.  Then
the DCT plus non-uniform quantization throws away both depth
and resolution of both color and luminance in something like
the frequency domain, given that some of the higher frequencies
are sometimes quantized to zero bits, depending on the
compression ratio.  etc.


IMHO. YMMV.
-- 
rhn A.T nicholson d.0.t C-o-M

Richard Owlett wrote:
> I understand that JPEG is a "lossy" compression method.
> 
> For color images is loss in:
>  1. resolution
>  2. color depth
>     a. would that include contrast
>  3. resolution AND color depth
> 
> I intuitively suspect #3.
> If so is it possible trade off resolution for color depth and vice versa?
> 
> If not, are there "high" [definition of "high" to be specified by 
> responder] compression algorithms which allow selecting preferred trade 
> off? I would *presume* something defined as a "quality factor" 
> [~proportional to resolution times color depth] would be held constant.
> 
> secondary questions
> Have I properly framed the question?
> Have I framed it well enough to demonstrate that I understand what DSP 
> is about even if I can't do it?
> 
> 
> 
My understanding is that loss in JPEG is primarily in resolution.  JPEG 
(at least before JPEG-2000) only specifies the decoding -- so the 
encoding can vary from one vendor to another.  The idea behind JPEG 
encoding is that you want to analyze the image for the amount of content 
and allow resolution loss where the image has fairly low contrast, but 
retain high resolution where the image has sharp edges or other areas of 
high spatial detail.

JPEG itself allows you to choose the amount of compression you allow -- 
so you can choose anything from totally lossless compression to an image 
that has a header and three bytes specifying the tone and tint of a 
single large pixel.

 From both a user-space and a technical perspective JPEG-2000 is a 
Really Cool Algorithm.  It doesn't seem to have caught on because there 
are some royalty issues holding it back, but it has some really nice 
features that'll make it unbeatable if the business end ever gets sorted 
out.

-- 

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

I understand that JPEG is a "lossy" compression method.

For color images is loss in:
  1. resolution
  2. color depth
     a. would that include contrast
  3. resolution AND color depth

I intuitively suspect #3.
If so is it possible trade off resolution for color depth and vice versa?

If not, are there "high" [definition of "high" to be specified by 
responder] compression algorithms which allow selecting preferred trade 
off? I would *presume* something defined as a "quality factor" 
[~proportional to resolution times color depth] would be held constant.

secondary questions
Have I properly framed the question?
Have I framed it well enough to demonstrate that I understand what DSP 
is about even if I can't do it?