MIZZI Computer Software

Understanding noise filters

   

To understand what noise filters can do let us consider the following image.
f/3.4,
1/8 sec,
ISO 100
click to enlarge it

First let us make three shots for ISO 100, 200, 400 and crop a region that fits on the screen. Now we compare the ISO 100, 200, 400 exposures for that small cut.
f/3.4,
1/8 sec,
ISO 100
f/3.4,
1/15 sec,
ISO 200
f/3.4,
1/30 sec,
ISO 400

You see that the images will get blurred and noisy to higher ISO values. Perhaps you assume that the ISO 400 image should be as sharp as the ISO 100 image and should have only more noise. You see this is not true.

Now let us compare the ISO 100 images with different noise filters
f/3.4,
1/8 sec, ISO 100
no filter
f/3.4,
1/8 sec, ISO 100
noise filter level 1
f/3.4,
1/8 sec,
ISO 100
noise filter level 2

You see that the filters are not producing any blurring nor they remove edges or small structures completely. Perhaps you can see that some structures of similar brightness will result in structures with lower contrast.

Now compare the ISO 400 images with the filtered images....
f/3.4,
1/8 sec, ISO 400
no filter
f/3.4,
1/8 sec,
ISO 400
noise filter level 2
You see that the noise filtered image has a less noise in dark regions but in regions where many structures are found the image seems nearly unchanged. As you can see in this sample: there are images where your eye finds that an ISO 400 image is better than it really is. There must be only enough structures to confuse your eye. But if you look to the right log you may also see that there are really strong changes in the colors of the log.

Now compare an other region of the ISO 400 image with the filtered image
f/3.4,
1/8 sec, ISO 400
no filter
f/3.4,
1/8 sec,
ISO 400
noise filter level 2

The clearly visible noise in regions without structures is perfectly reduced to nearly the level of the ISO 100 image, but note that in dark sections some contrast in the structures are lost.

What we have learned: Noise removal is more visible if you have images without too much structures. Let us again compare our example image:

First the original one....

and then with a noise filter at level 4 (less than in the sample before) ....

The noise reduction level was lower than 2 and even so the reduction seems to be higher than in the examples before with reduction level 2. Your eye is really not a perfect one. Noise reduction is applied in regions where not to much structure is present, but these regions are most important because your eye is sensitive to noise in this regions only. Blurring and smooth edges produced by higher ISO levels can not be reduced or improved by noise filtering methods.