How Do I Measure Noise Using Photoshop? And What Is A Significant Change?

Started 2 months ago | Discussions thread
John Sheehy Forum Pro • Posts: 22,666
Re: How Do I Measure Noise Using Photoshop? And What Is A Significant Change?
1

MrBrightSide wrote:

Question 1: Long ago I remember reading that you could use the statistics tools in Photoshop to measure the noise in a photograph.

Measuring noise in a converted image has no direct relevance to the underlying RAW noise, because you do not know how sharpening, NR, and demosaicing have distorted the noise vs frequency spectrum, and the gamma or tone curve applied distorts what the actual signal level is. All you can measure is the the result of all processing, as a statistic with very limited meaning.

Measurement is best done on non-converted RAW data. You can get the RAW data into Photoshop by having it converted to a 16-bit greyscale image by IRIS (for older cameras), dcraw (older cameras), RAWDigger (more up-to-date), or libRAW's "unprocessed_raw.exe".

There must be something in the program because it has noise reduction so it must be able to identify noise. How does it work?

If software could identify noise, that would be wonderful, but for the most part, all it does is guess or eliminate outliers including potential real dot-like signals, and totally miss some of the most visible noise, like banding noise, assuming it to be banded subject matter.

Failing that, what other simple methods are there to measure noise in digital photos.

As much as I like math and measuring things, I much prefer to use DPR's studio comparison tools these days. They do have some variation across cameras in sharpening, contrast, and NR, but you can learn to mostly see through these things and see the noise in a spectrum.

Question 2: Once the noise is measure what constitutes a significant difference in the noise level? Does it fall at 1 percent noisier? 10 percent? 100 percent?

That depends on threshold phenomena. If we start with a synthetic, computer-generated image with no noise, we can take one copy and add 1 unit of noise, and add 10 units to the other (10 times as much), and you may not see any difference at all. However, if you add 1000 units to one, and 1100 units to the other (only 1.1x as much), you might clearly see the difference, in an A/B comparison. It isn't until noise levels get above the threshold range that perceived noise gets somewhat proportional to empirical noise.

Below a certain level, depending on context (including sharpening and image display size), lower noise can not be appreciated, and in fact, too little can actually be a bad thing in smooth gradients, which will posterize into contour banding with too little noise.

Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow