Digital BW, The Print

Roy, 

Thanks for your comments on my write-up. You raise good points, many of
which were noted by others the first time around.

Your comments on the subjective nature of the challenge are all extremely
valid. Unfortunately I think the challenge was concocted to to get the ball
rolling, to induce people to get involved in the conversation, to show their
images and results; but in the end people just got too hung-up on the
specifics of it to play. They took a "challenge" as a fight, and chose to
stay away. But it seems even a $100 "reward" doesn't do the trick either, so
there's just no way to win. Anyway, those challenges are for color images
and we are here for BW, but the same subjective conditions do apply. I
simply did my piece to take up the call to demonstrate some results through
images, even if it's not an entry into THE challenge.

With that said, I put a lot of energy into trying to define the scenario in
my piece and not enough into images. I could post 100% TIFFs and will try
to, but somehow I trashed my original images, so I'll have to regenerate
them, which is what's slowing me up. Ugh, another project...

The thing is, as you say, all images will respond differently, and my hope
really was that some others might be inspired to take their own images and
do something similar. Then we can get a sense of what may happen across a
broader image base. I think that would be better than getting bogged down in
my one image. I fear I'll never choose an image, select a crop, or
manipulate the file, etc, to everyone's liking, so we'll see. Maybe my piece
is done as is and someone else can do it better...

As for including a grayscale image, the inestimable Antonis also suggested
that, so I may try to include that if I do a revision. Here my fear is that
people will fixate on a posterized gradient, without regard to how grain,
detail, noise, etc, - IOW, the components of an image that differentiate a
photo from a pure gradient - mitigate the effects of posterization. Most
beginners believe that blurring and smoothing an image will mitigate
posterization, when in fact it's the opposite - adding noise (randomization)
is the fix. That's why PS adds dither when converting between modes. That
dither IS one reason that images sometimes look better than we think they
should.

You also listed several factors, notably image downsampling and JPG
compression, as factors that could have made my images look better than
expected. This raises a question from me. If you have the time to test this
I'd be extremely interested if you would try to replicate those processes on
your pure and posterized gradient images, and either post them or describe
the effect. That might go a long way toward confirming or denying those as
relevant factors.

Thanks Roy,

Todd



on 5/25/02 12:38 AM, royvharrington wrote:

> Hi Todd,
> 
> I'm kind of a late comer to the discussion, but I missed your
> article first time around.  I went back and found it, read it and
> have a couple ot things to add.

> First of all, I was reluctant to get into the 8 bit challenge because
> so much of it is subjective and unspecified.  What's a
> reasonable negative to be used, what's reasonable usage
> of the scanner, what's not too much tonal manipulation and
> how visible does the posterizing have to be.  Most people
> could come up with a fabrication of posterization but then
> there's always the argument about whether the process
> was "reasonable" i.e. could it happen in real work.
> 
> Todd's article at least is a reasonable place to start.  He shows
> a fair amount of tonal manipulation in 8 bit mode and shows
> results that at least on the surface are surprisingly good.
> 
> There are several factors that I think may make the result look
> better than it expected.  I can't really be sure these would
> make a difference but maybe Todd can tell from the originals.
> 
> 1)  To make the images fit in the article, they were shrunk
> to 50%.  I assume this was cropping and resampling down
> in PS.  This means that 4 pixels in the original where averaged
> into 1 pixel in the output.  By doing this, the gaps are effectively
> filled in partially.  If there was a jump in the histogram from
> 100 to 120 its possible that 105, 110, and 115 levels were
> generated by the averaging.
> 
> 2) In the html of the article the images are again shrunk
> slightly by about 82%.  This means the browsers are again
> resampling and averaging all the pixels.
> 
> 3) Compressing with jpg also slightly changes all the levels.
> The uncompressed files are never identical to the original.
> 
> All these may seem small individually, but as was observed
> if you read the jpg's into PS and look at the histograms, they
> are completely filled up except for going all the way to white.
> 
> ---------
> 
> The next issue I think is the choice of image and cropping.
> The image is great for showing some need for tonal
> manipulation, while not overly drastic the changes to create
> some combing in the histogram.  The other major issue is
> that posterization is only going to be obvious in a very
> gradual gradation of tone.  Detail like hair is great for sharpness
> but won't show posterization no matter what you do to it.
> 
> So here's my idea:  Let's keep the picture and the tonal
> corrections, but let's add a slow gradient at the bottom.  I took
> the original jpg from the article and added a PS gradient
> from 40%K to 0%K below.  I just did the light end since
> the corrections make it darker and also the posterizing is
> likely in the white end.
> 
> Now I applied the tonal changes that Todd had in his article.
> Well sure enough the gradient shows very obvious posterization
> right on the screen.  This was all done in 8 bit mode.
> 
> I figured I ought to try it in 16 bit mode.  First problem is that
> the gradient tool doesn't work in 16 bit mode.  So I put the
> gradient in in 8 bit mode, converted to 16 bit, and then ran
> a gaussian blur with radius 4.  This put at least some
> data down into the lower bits, but I know of no way to "see"
> what those lower bits are.  Rerunning the tonal changes
> there was no posterization with the 16 bit mode and the
> histogram was jagged but pretty much filled up.
> 
> ----------------------------------
> 
> So, does this prove anything?  Probably depends on what
> you want to think.  I tried to show that what might look OK
> could potentially also result in problems.
> 
> For me, I figure that the raw scan must be done in greater
> than 8 bits (12 to 16 or so), most tonal changes should be
> done in high bit mode but once they are done conversion
> to 8 bit is OK before printing.  This conversion can be done
> in the scanner software and just pass on 8 bit to PS, or
> we can pass all the high bits to PS and delay the conversion
> as long as desired.  You just have to decide when "most"
> tonal changes are finished.
> 
> Roy