Denoise

[William Dickson]

I recently bought a Canon 5D Mkiii after using the 7D. I had lots of noise issues whilst using the 7D, but, since purchasing the 5D Mkiii I have discovered that I can now use a high ISO and still get decent IQ.

I captured the attached image with my 5D Mkiii, Canon f/2.8 300mm with 1x4iii attached.

f/4
1/2500
ISO 3200

It is a big crop, I used Topaz Denoise 5, and sharpened the Robin slightly, I intentionally used the high ISO setting.

The weather was fairly dull at the time.

What do you guys think on this issue.

[Ian Wilson]

Hi William,
You raise a very interesting subject which I suspect has been discussed many times before in this forum. Recovering signal from noise is a huge subject and I am no expert but I suggest the efficacy of the various techniques boils down to (a) how faithfully the recovered signal resembles the original and (b) the amount of residual noise. I would be happy if the recovered signal appeared to be artifact and noise free when output on the device for which one prepared the image. In the case of your example, a 366 kB file 1024 x 682 pixels, prepared for viewing on a computer screen, (a) and (b) are reasonably well achieved. There are no obvious artifacts, though I suspect there is some loss of fine detail, and there is no obvious noise problem. Overall the image does appear a bit 'soft' which could be a combination of (a) and (b) or it might just need a bit more sharpening.

It would be nice to see a comparison of the performance of various NR products using a set of 'standard' test images with various amounts and types of noise. I have played around with noisy grey images which enable one to quantify the mean (signal) and standard deviation (noise), and hence the signal-to-noise ratio, but from such a simple test case it is not easy to get a handle on what would happen to the spatial frequency content of a real image. Perhaps one of our forum members can tell us more.

Regards Ian.

[Randall Farhy]

Hi William, overall the picture looks clean, especially for this ISO. I'm wondering though-did you mask the bird and only apply noise to the BG, or was the NR global with selective sharps on the bird? I only ask because the technique appears to be the latter as the bird is showing some softening in the finer details which could be salvaged with selective NR on the BG. If you did use selective NR on this image, then the crop factor might be the culprit. It's minor, and depending on the size of the crop-probably a non issue for most images. TFS.

[William Dickson]

Hi Randall

I only denoised the BG and applied USM at 40 to the bird, but, yes it was a very large crop

[Diane Miller]

As with just about everything in digital processing, the best fix for any problem is prevention. Keep the ISO as low s possible and, above all else, don't underexpose critical areas (areas where you will want to keep detail). This post looks OK to me, but I'm not into enlarging in the browser or putting on reading glasses to peer at the dot pitch of my monitor.

You can be the best judge, viewing the corrected file at 100%. Different images may need different treatments. As Ian said, cleaning up noise without loosing fine detail can be very difficult as their spatial frequencies can be very close.

I've had good results with Nik's Dfine, Neat Image and Topaz DeNoise. Sometimes one work better than the others. If I'm careful to keep the original exposure up, I rarely need to mask the subject when I want to denoise the BG.