Interpolation and the hidden Unsharp mask.

[Greg Vander Houwen]

Sir Rick mentioned doing an Unsharp Mask to images before/after interpolation here. (http://photoshoptechniques.com/forum/showthread.php?postid=15887#post15887) It reminded me of something I wasn't aware of until recently.

An Unsharp Mask or USM is applied by default when you do an interpolation (Image Size, Transform...). So, although a USM may improve the image after interpolation keep in mind that you are already running a USM automatically when you do (any?) interpolation. There is probably an exception to this.

Note: I haven't tested this on any method other than Bicubic. I expect it would be applied with Bilinear as well but probably not with Nearest Neighbor.

In the figure below the lower right section is the original image at 250 x 250 px and the larger background is a detail of the same image after an Image Size, Bicubic interpolation to 2500 x 2500 px.

Greg

[theGuest]

Hey cool info guys... now that's something i didn't know either.

I'd still keep in mind though that PS is probably just applying an all-purpose, generalized sharpening, and not a specific one. Which may still be needed after scaling, to enhance/replace any lost details in specific areas of your photo/image.

Thanks for the info Greg. }$

[Greg Vander Houwen]

Yes, Sir Mark, good point.

I haven't tested this in detail as to how/if it determines the amount of sharpening but what ever it does there is no UI control.

This is something I have been meaning to mention to the engineering team. It might be useful to have control over although I don't know how you would provide a preview. Or as another option to have a switch to shut it off so you could do the USM once but with the standard UI.

Always stuff to learn... life's good ;).

Greg

[Rick Miller]

Additionally, I'd like complete control over the lightening and darkening of the sharpening halo so I can make modifications to them independently. Linocolor elite software from Heidelberg for the Linotype-Hell (also for Umax) allows one to target just the light contour or the dark contour.

Could ya' put in a word Greg?
;)

[markzebra]

Thats a new one on me too.

So does it apply USM to both upsampling and downsampling?
If so does it do this equally?
Is there some kind of formula at work?


$}

Rick hmmm .. I think there may be a way to isolate the light halo from the dark - using Difference and Lighten or Darken on a duplicate layer?

[Greg Vander Houwen]

Good questions Nessie ;). Dunno.

I'll do some asking around and see what I can come up with. In the mean time if any of you do some testing on this, please post what you discover.

Greg

[Rick Miller]

I think there may be a way to isolate the light halo from the dark - using Difference and Lighten or Darken on a duplicate layer?
Yes, Dan Margulis illustrates this in his Professional Photoshop 6. I just wanted to have an easier and more convenient way of doing this.

So does it apply USM to both upsampling and downsampling?
It appeared to when I briefly tested and measured the numbers with the info palette. Will have to look more into this.

[markzebra]

What exactly is Margulis de Sade's technique for this Rick?

I was just thinking I guess of splitting using one Lighten layer and one Darken layer. The Difference thing was as a way of getting the sharpened pixels to separate on a layer of their own - probably not necessary now I come to think of it

[Rick Miller]

What exactly is Margulis de Sade's technique for this Rick?
The exact method you just explained. Creating two duplicated layers - one set to Lighten, the other set to Darken.

[theGuest]

YES!
I agree with Rick! I too would like full-on control like that.

Good thinkin Rick!}$

[Raphael]

At least being able to turn it off would be good.

Check box maybe?

}Raphael{

[sPECtre]

For those starving for technical data and the real Why, here is some info from Gernot Hoffman:

Pierre,

in test images with sharp edges PhS Bicubic creates some blur
and additionally halos. Now it depends whether this is perceived
as more blur or as an illusion of sharpening.

Nothing without proof (if somebody wants to have a look at this
1.1 MB file), page 3, center:
http://www.fho-emden.de/~hoffmann/bicubic03042002.pdf

Essentially, this effect can be explained by a simple example:

(*) original and interpolated values
(+) interpolated values

-------------------------- *
--------------------------+
-------------------------+
-----------------------+
-----*---------------*
--------+----------+
--------------+

The negative overshoot creates the halo.

Bilinear doesn´t have overshoots, therefore it´s sometimes better
for line art.

Best regards --Gernot

[sPECtre]

Follow-up:
------------------
Pierre,

my link (unfortunately very large) should show some results
for different algorithms. In this Forum the mathematical part
can be simply ignored. The issue was discussed extensively
in the Google Computer Graphics Forum.
There are lots of "ideal" interpolation schemes, but some of
them require float accuracy and are very slow.

Therefore I had focussed my "research" on four types:
Bilinear (similar to PhS)
Biquadratic
Bicubic (probably like PhS)
Bicubic spline (no overshoot but blur) + sharpening

I had posted here, because I thought it might be interesting to
explain this contradiction blur/sharpening.

Yes, Bilinear connects the given points (asterisks in the ASCII
drawing) by straight lines. No overshoot.

Yes, please quote my URL. I am open for any objection and
correction, if somebody should find bugs.

Best regards --Gernot

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

If someone has the mathematical skills, it might be useful, but I think I need to go back to school 8(

[john opitz]

I was doing some testing of sharpening actions and other actions for Steven Marsh,using convolution. You might want to check them out.

Binary FX (http://members.ozemail.com.au/~binaryfx/PSTV_convolcorner.html)

John

[Greg Vander Houwen]

When I was at the i3 forum I spent some time talking to Mark Hamburg over dinner. He was the lead architect for PS for many years and just recently moved on to other projects within Adobe. I asked him about this as he knows PS (technically) better than anyone I know. He pointed out that to describe this effect as an "Unsharp Mask" was misleading because there is no user control.

Greg