Re: TIFF vs JEPG

On Fri, 21 Apr 2006 00:44:18 +1200, Colin D
<ColinD@xxxxxxxxxxxxxxxxxx> wrote:

I loaded both of Rafe's images, the tif and the jpg into Photoshop,
They are both exactly 1001 pixels square, so if scanned at 4000 ppi the
crops represent a 0.25-inch square piece of the negative or slide, and
when viewed at 100% on my monitor at 72ppi, the magnification is
4000/72, or 55x.

Without getting into details, for an objective test you have to scan
raw, in other words disable everything. One common trap often missed
is interpolation. There are many such pitfalls which is why testing
parameters are so crucial. This goes for any objective test.

Then I made a duplicate image in PS of the tif file, and saved it as a
level 12 jpg, wrote it to disk and then reloaded it into PS.

OK. Results: at 400% on the screen, each pixel was visible, and at that
level there was no discernible difference among the three images side by
side

There are many reasons why that's not a good procedure (the same goes
for layers) but I won't get into any of that now. Instead of side by
side, open both images independently. Next, click on the magnifier and
set the checkmark "Resize windows to fit". Finally, double-click the
magnifier for each to blow up to 100% and - if the images are larger
than the screen - this will also automatically align and overlay them.

NOTE: Photoshop (at least my version) may be cranky when lining up
images using the above process. This alignment will *only* work if you
haven't changed image size after you opened them! There's a workaround
but the easiest thing is to just do exactly as I explained above.

After that use Control/Tab to flip between the two images. If you
scroll them both by the same amount (scroll, Control/Tab, scroll),
they will remain aligned. Look for "lines" i.e. sudden and distinct
borders between neighboring areas. That's the JPG 8x8 square.

Too much magnification may be as bad as too little. I find that (for
all purposes actually!) ~300 magnification seems to work the best.

Under blind test
conditions I don't think anyone could nominate which was which.

As I also mentioned even if everything was done properly up to this
point a lot depends on image content. If you're looking at "black cats
in tunnels" or "polar bears in snow" you won't see much.

But even if you're looking at an image segment where JPG artifacts are
most pronounced it's still a subjective call unless you know what to
look for. But to a trained eye JPG stands out like a sore thumb.

An interesting point, though. I understood that for the best jpg
results the pixel dimensions should be divisible by 8, which 1001
certainly isn't. Since the jpg's remained at 1001 pixels, the jpg
process didn't crop the image, so maybe it ignores the extra pixel, and
just operates on the 1000 x 1000 area.

Anybody know?

That's not quite correct. JPG itself works by dividing the image into
8x8 squares and compressing them individually. That's why the border
between two neighboring squares is visible. In general, the algorithm
only takes into account pixels within that 8x8 square. The square next
to it is "on its own" which is why there is a border. If the image is
not divisible by 8 there are a number of strategies to compensate for
the missing pixels but that should not affect the rest of the image.

Don.
.

Relevant Pages

  • Re: 16:9 Woes
    ... area (which is 702 pixels wide) in the middle. ... I assume from that article that these are square pixels, ... them _might_ have video clip recording modes that do not. ... then resample to 720×576. ...
    (rec.video.desktop)
  • Re: Monitor aspect ratio
    ... desktop resolution stretched). ... given that his screen's square pixel ... is at an aspect ratio of 1.6, then you can calculate the ratio you need for ... pixels, then his whole Windows experience is stretched, and the square ...
    (microsoft.public.win32.programmer.directx.graphics)
  • Re: Hashing
    ... > use in a compiler/assembler. ... It shows a square of Pixels, ... Algos, the easy way: The better the Algo is, the more ...
    (alt.lang.asm)
  • Re: CD/M2
    ... the pixels were ratio 2:3 - made drawing a circle interesting as you had ... but these were definitely square. ... finding a similar size widescreen CRT is becoming more and more ... mounting machine screws aren't M4 like in the VESA specification. ...
    (uk.tech.digital-tv)
  • Re: Crop vs Resize
    ... pixels are ... There is no loss of quality. ... I'm shooting Nikon and actually work on .dng files. ... but not just the .jpg. ...
    (rec.photo.digital)