Aliasing and supersampling, why small pixels are good.

Friday, 20 December 2013 12:46 administrator

With newer technology pixels tend to shrink. No doubt, megapixels are a selling argument. But there is a very good reason to reduce pixel size. Small pixels are the best way to defeat aliasing artifacts.

Check the image below (it reduced in size. Click on image for original size):


These images were shot at 80 mm focal length at 4 m distance with a Hasselblad 555 ELD using a Planar 80/2.8 at f/5.6, f/11, f/22 and Sony Alpha 77 with a 70-400/4-5.6 G lens at 85 mm and f/8.

The discussion here is not about MFD vs. smaller formats, it is about pixel sizes. Both images are about 1:50 scale. 

What is quite visible, even on downsized image above, that the P45+ images have colour artefacts at both f/5.6 and f/11. The colour artefacts go away at f/22 (due to diffraction) but the images also lacks resolution and gets softs. This is due to diffraction. The image shot with the APS-C camera is clean and has best detail.

Look at the image below (click on image for uncropped version). Both images show Norman Koren's test chart at 1:50 scale. Top one is from the APS-C camera. The theoretical resolution of that camera is 128 LP/mm. The test chart shows that contrast is low at the limit of resolution, known as the Nyquist limit. There are some artefacts, but they are quite weak. This i a well balanced behaviour. The lens probably reaches it's resolution limit at the same time the sensor does. There is little contrast in the projected image at the Nyquist limit. Few artifacts arise. 

The bottom image shows the much larger pixels of the P45+. Here the Nyquist limit is around 72 lp/mm. The Planar lens has plenty of contrast to give at 72/lp mm, so there is a lot of detail beyond the Nyquist limit resulting in a lot of artefacts. These artefacts are also colorful, due to Bayer RGBG colour grid array and lack of optical low pass filter. The sensor does not resolve past the red line, so any detail to the right of the red line is fake. This obviously doesnt apply on the fixed frequency bars that are added for reference.

If we check out the text below the Norman Koren test chart we can read some fine print. The text is clearly resolved on the Alpha 77 but show color pixels on the P45+, stopping down the Planar 80 on the P45+ to f/22 removes the color aliasing at the cost of loss of acutance due to diffraction. The image below is shown as a TIFF as JPG compression removes some of the fake colours. 

What we can see from the examples above that false information will arise if the lens ouresolves the sensor. This is quite obvious, if the lens has higher resolution than the sensor detail will jump between pixels, transition cannot be smooth.

Feather shots

To demonstrate that not only test charts are affected, I photographed a feather. It is probably coming from a seagull, I just picked it up on a beach. 

The two images below were shot with two different cameras with 150 mm lenses on both and at 4 m distance. The reason the Sony image is larger is that it has smaller pixels. Would the P45+ MF back have the resolution of the Sony Alpha 77 APS-C chip it would have something like 118 MP. So small pixels on large formats make a lot of sense to me.

The feather showed a great amount of aliasing artefacts. I have taken the bottom part that shows a cross hatch pattern instead of single strains. I can mention that I got this kind of artefacts both on a Planar 80/2.8 and on the Sonnar 150 at the same shooting distance. Stopping down to f/11 was not enough to reduce them significantly.

P45+ Sonnar 150/4 at f/8Sony Alpha 77 70-400/4-5.6 at 150 mm and f/8P45+ image enlarged for better viewing in Photoshop 

The next pair of images are the same as above with Sony image reduced to approximately the same size as the P45+ image.

P45+ Sonnar 150/4 at f/8Sony Alpha 77 70-400/4-5.6 at 150 mm and f/8

Test 3 the Dollar shots

I frequently use a One Dollar bill for testing. It is available everywere and has a lot of fine and nice detail.

I have shot a series of images with a Planar 80/2.8 at different apertures and also shot with an 16-80/3.5-4.5 zoom on my Sony SLT 77 at f/8. 

The images below are intended to show the difference betweel large pixel capture and small pixel capture. They were shot with lenses having similar focal lengths and same tripod and subject position. Image size and pixel count is ignored in thi context,

Lets compare two decently sharpened images. The left one is coming from the P45+ and the right one is from the Alpha 77.

P45+, large 6.8 micron pixels SLT 77, small 3.9 micron pixels, downscaled
The P45 image has moiré around the Washington portrait and low frequency aliasing artifacts within the red frame The SLT 77 has much less moiré and artifacts

The Alpha images was downscaled, the real size image is here, note that it resolves the fine lines within the red frame, although having some broken line artefacts. Some false colour is noticable but most of the moiré around Washington is gone.

The figure below shows the MTF plots of both images without sharpening applied. The Nyquist limit on the P45+ is around 73 lp/mm while on the SLT is around 128 lp/mm. We can see that MTF at Nyquist on the P45+ is around 10% and on he SLT 77 it is around 3%. The sensor cannot resolve beyond the Nyquist limit, so any detail beyond that limit will be shown a artefacts, called aliases. Some of the aliases will be colour artefacts.

Sharpening will exaggerate the artefacts, check the figure below, here the images were sharpened a bit aggressively. 

No You See It and Now You Don't

Stopping down from f/8 to f/16 reduces resolution (and MTF) due to diffraction and nearly eliminates aliasing at a cost of significant loss of sharpness. Some of the sharpness can be restored by more aggressive sharpening.
Full size JPEG
Aperture f/8, weak sharpening. Ver obvious artefacts.
Full size JPEG
Aperture f/16 quite a bit more sharpening. Most artefacts are gone.

The small pixel advantage

It seems that reasonably small pixels, like 3.9 micron used in this example resolve significantly more detail than larger pixels. A major benefit of smaller pixels is that they produce less artefacts.

Real world examples

This image shows "staircase aliasing" on the near vertical wires of the sailboat rig. The same wires show color aliasing. The bright leaves in the background also have color aliasing.

The above image shows color aliasing when small subject detail goes across several pixels. It is well possible that it goes away in print. Just converting to JPEG makes it less obvious, as JPEG smears colours a bit.

The final example show colour aliasing on water surface. Moiré filtering works on this but needs to be applied locally. Even in monochrome it gives a rough surface on water.

So what did I see?

It seems that aliasing and speciall colour aliasing can be a real problem in MFD. The reason for this is that MF cameras may have relatively large pixels with no OP filtering and lenses of quite decent quality. I don't know why OLP filters are not used on MFD backs. The OLP filter itself is pretty expensive and that may play a role (the Mamyia ZD had a removeable OLP filter with a price tag of 3000 $US). The main reason I guess is that per pixel sharpness looks good without OLP filtering.

Sharp lenses, fat pixels and no OLP filtering is a recipe for aliasing, and alot of it can be seen in my images, enough to cause some discomfort. But, in most pictures it is not obtrusive. Why? A reason may be that color aliasing or aliasing in general can often be seen on simple structures and on repeating patterns. The detail or pattern must also be close to or below pixel size. These conditions are not met very often. 

An example may be the One Dollar bill at the start of the article. It shows a lot of color fringing but little detail. It was shot at 4.0 meter with an 80 mm lens, a 1:40 reduction which is pretty decent for lens testing. For the moiré comparison I moved in close around 2 m (?) and got even more moiré. A close up would show no moiré as the dollar bill would not have fine enough detail to cause moiré.

Also, moiré is actually an indication of good technique, it indicates good focus and appropriate aperture. It seems that stopping down to f/16 on a 6.8 micron pixel back virtually eliminates aliasing artefacts.

What I found is that reducing pixel size makes a lot of sence. The image may be a bit soft with reduced pixels size but a sensor with smaller pixels will always record better detail than a sensor with larger pixels.

The latest generation of MF backs use 5.2 micron, they are normally regarded to give superior quality to older backs like my P45+. That image quality comes with a price tag, however.
Last Updated on Wednesday, 19 February 2014 13:40