" />

5D MARK II - “REAL” NOISE TESTS

The photo used in the samples below was taken a week and a half ago on the Washington Coast. The moon was about first quarter and the clouds were really thick so there really wasn’t a lot of light. I wanted to at least get a few images done that night before the moon set so I used ISO 400 and exposed most of the images, including this one, for 15 minutes with Long Exposure Noise Reduction turned ON.

All of the images could have used at least another stop and a half of exposure - the histograms barely breached the 50% mark. My estimates based on a preliminary ISO 6400 exposure should have been better but I think the light was lessening slowly as the moon set which I didn’t account for. While this isn’t a perfect test scenario I decided to use it because it was a real world situation where although I didn’t correctly expose the image I still really like it and it’s certainly an interesting test of the camera’s capabilities.

The image was taken in portrait position but for this test I cropped off the top and bottom (mostly sky and sand). The first image below is a resized version of the image prior to any adjustments to the histogram and the second image is post adjustments (listed below the image).

Original unprocessed image in DPP 3.5 - click white balance, no adjustments to histogram using RAW or RGB settings, sharpness 0, luminance noise reduction 0, color noise reduction 5, resized to 500px wide

Processed image in DPP 3.5 to stretch the histogram to a “better” exposure - click white balance, brightness +2, contrast +4 in RAW, no adjustment in RGB, sharpness 0, luminance noise reduction 0, color noise reduction 5, resized to 500px wide

Here’s the histogram before and after the DPP adjustments. As you can see it’s a significant adjustment and as well all know an adjustment of this size will dramatically increase the noise levels in the image.

[Warning! The numbers you will read next are estimates only and not intended to be exact.] For an image shot originally at ISO 400 and pushed the equivalent of 1.5 to 2 stops it makes the effective ISO somewhere in the 1200-1600 range. Assuming this is in fact the equivalent of a push from 400 to 1600 there will be more noise in this image after adjustments than there would have been if I had correctly exposed the image with a 15 minute exposure at ISO 1600 in the first place.

What I’m really getting at is that this image should be damn noisy!

Not much to see yet so here are 100% crops from the image. The first in each set is a crop from the original image and the second is a crop from the adjusted image. The first set has one additional crop which is the crop of the adjusted image with the Dust & Scratches filter applied (see my last post). Keep in mind that with the exception of the third crop from set one no software based luminance noise reduction has been applied.

Crop 1

Original

Adjusted in DPP

Dust & Scratches

This shows pretty much what I would expect - decent noise levels in the original and increased noise levels in the adjusted image. The Dust & Scratches filter definitely makes a difference in removing some of the speckles and results in an improved image. Overall the noise levels really aren’t bad at all.

Crop 2

Original

Adjusted in DPP

Crop 3

Original

Adjusted in DPP

Very similar results for crops two and three - increased noise levels in the adjusted image but still holding up very well considering the circumstances.

And finally, crops from a screenshot taken at 50% viewing (print size was 40%) which are similar to what you would actually see on a print.

50% crop 1

50% crop 2

I don’t see anything in the 50% crops that would prevent me from getting an excellent print from this image especially considering I have made absolutely no effort at noise reduction or selective sharpening.

So while the 100% crops are moderately (but not unexpectedly) noisy (remember it was a 15 minute exposure at ISO 400 pushed the equivalent of 2 stops!) the 50% crops look great and in my opinion it’s an impressive performance.

[Update] I should add that once the Pandora’s box of noise reduction and sharpening has been opened the results are so wide ranging that it’s probably impossible to definitively conclude what is the “best” result without actually seeing a final print (no one in their right mind would publish a full size image on the web for “final” viewing). Since I published this post I’ve been playing with a variety of noise reduction and sharpening tools/scenarios and while it’s clear this image can be vastly improved it’s hard to say which tools/scenarios yields the “best” results. However, so far I would venture to say that once printed (excluding massive enlargements and assuming a well prepared image and a knowledgeable printer) no one would even mention the noise as an issue.

Leave a Reply


Comment moderation is in use. Please do not submit your comment twice -- it will appear shortly.