HowTo: subtract filtered image from orig and calc stddev?

look for ready-made device noise profiles and offer your own ones here
Post Reply
l_d_allan
Posts: 35
Joined: Thu Sep 24, 2009 3:00 pm

HowTo: subtract filtered image from orig and calc stddev?

Post by l_d_allan »

I was browsing older posts, and came across this one:
http://www.neatimage.net/forum/viewtopic.php?t=1361

In that thread, Vlad wrote:
> To estimate the noise level of the processed image, subtract the filtered image from the original one and then calculate stddev for the result. The higher noise levels are, the larger stddev will be.

This seems interesting, but I can only speculate how to do such a subtraction. Is the procedure described somewhere?

I suppose for a 10 mpx camera, this might involve comparing/subtracting the R, the G, and the B values for 10 million pixels, and calculating the std-dev of the differences?
NITeam
Posts: 3173
Joined: Sat Feb 01, 2003 4:43 pm
Contact:

Post by NITeam »

Yes, that would involve such kind of calculations. Can be done in Photoshop or any other advanced editor.

Vlad
l_d_allan
Posts: 35
Joined: Thu Sep 24, 2009 3:00 pm

Post by l_d_allan »

Is the procedure described anywhere?

There are several reasons I'm interested in being able to quantify noise values. One is to compare different cameras at different ISO. Also to check if ETTR makes that much difference (see thread http://www.neatimage.net/forum/viewtopic.php?p=7700)

The other is to try to evaluate how well different NR tools work, specifically if NR in LightRoom 3.x and ACR 6.x have gotten better than LR 2.x and ACR 5.x

I have PS-CS4 with ACR 5.x, and was disappointed with the NR accomplished by ACR 5.x. That was the justification to purchase NI.

In the marketing material for LR 3, Adobe indicates that NR has gotten significantly better. I tried the original beta for LR 3b1, and didn't see all that much of an improvement as far as NR was concerned. I didn't try the subsequent LR betas, but I have read posts that NR got better. At some point I'll download the CS5 trial, and one of the things I'll evaluate is whether the NR is better.

My speculation is that NI is still better than the NR done by ACR 6.x, but it would be useful to have actual numbers. It seems like the std-dev might be the equivalent of a "figure of merit" to do meaningful comparisons.
NITeam
Posts: 3173
Joined: Sat Feb 01, 2003 4:43 pm
Contact:

Post by NITeam »

I am afraid quality of noise reduction is a bit more complicated to evaluate than just comparing the std-dev. For example, Gaussian blur can provide the best std-dev but fail on other counts.

Regarding using Photoshop to measure std-dev, the procedure is not described on our website, but it is pretty basic and you can find the details in Photoshop documentation.

Vlad
Wolverine@MSU
Posts: 16
Joined: Thu Jul 27, 2006 3:11 pm
Location: East Lansing, MI

Post by Wolverine@MSU »

Bring a picture (not noise-reduced) into Photoshop, duplicate the layer, select the duplicated layer in the "Layers" window, apply NeatImage to the duplicated layer, change the layer type from "Normal" to "Difference, Look at the "Info" window to see the Std. Dev.

May not be exact, and the precision is only to 2 decimal places, but can give some indication of NR amount.
l_d_allan
Posts: 35
Joined: Thu Sep 24, 2009 3:00 pm

Post by l_d_allan »

Thanks for the reply. Looks like it is close to what I was asking about.

After the steps you described, I have what appears to be a completely black layer showing.

However, with CS4, my Info Window doesn't show Standard Deviation. It has document size, size, scratch, efficiency, etc.

Am I looking at the wrong Info Window? Or did I leave out a step, or do something wrong?
Post Reply