Pictus wrote:
And some script as Photons 2 Pixels Images suggested ?

the script will be strictly on PS side, so nothing needs porting on IM side

deploying it as a "plugin" with integrated IM is a whole different thing

it will probably remain possible only to those who are savvy with computers
... unless the plugin is written in C+ with the IM libraries (which are supplied)

Pictus wrote:
It is the gamma factor playing tricks again
Here the luminosity channel from a gamma 1 blur average.

The whole gamma factor is here http://www.4p8.com/eric.brasseur/gamma.html

Guys, we have to investigate the gamma factor for the “high pass sucks”...
I bet the gamma is the sucker...

I'm still not sure about the added value of the "gamma factor". The example they give fails the resampling as designed, but if you adjust the values in the example to linear gamma, it will fail in exactly the same way using the gamma correction, and will give good results if you do it without. ... I hope you get my drift.

the perceived luminosity aside, the "off" feeling I get is that the wall looks more mortar and almost no bricks after the sharpening (which is clearly not the case in the original)

about the gamma and freq separation... it will be hard to work with, but if you add a corrective gamma curve as adjustment layer on top, it may work

Pictus wrote:
Point taken, but the are 2 different things, one is what looks better and another is what is more realistic.

BTW, Blur-aware downsample algorithm
http://www.cs.ubc.ca/~heidrich/Papers/EG.11.pdf

reality is subjective


interesting, thanks for the link

NothingIsRealButTheGirl wrote:


Filter > Blur > Average

on the image as I found it. Seems like the upper one to me.

This is actually a mathematical quirk.  Because you're doing the averaging in a 2.2 gamma space, the upper agrees.  Converting the image to linear sRGB first and then performing the same action has the opposite effect - the lower becomes 'correct'.

That said, more interesting to me is that when one moves to LAB (a gamma-independent system) and performs the same test, the top one is once again the more consistent with the original image when averaged.

As to gamma + frequency separation, leveraging gamma shifts in performing FS can have certain advantages depending on whether one is attempting to enhance local contrast (as in the original gamma argument's USM tests), or whether one is using FS for corrective purposes.  Generally the latter is going to be more intuitive in a higher gamma (1.8 or 2.2) whereas the former might be used for the sake of aesthetics (depending on one's POV and interpretation of the LAB results above).

And all that said, I do completely agree that Lanczos is and has been sorely missing form PS's repertoire.


... and sorry for the late entry into this.

Sean Baker Photo wrote:
This is actually a mathematical quirk.  Because you're doing the averaging in a 2.2 gamma space, the upper agrees.  Converting the image to linear sRGB first and then performing the same action has the opposite effect - the lower becomes 'correct'.

I just tried Edit > Color Settings > Blend RGB Values Using Gamma 1.0 to see if it would change anything.

It didn't seem to affect the way the Average was calculated.