Aliasing and fake detail

Sunday, 12 May 2013 11:11 administrator

This article is about aliasing and fake detail.

Here are some recommended pages on the issue:

Bart van der Wolf article 1 and article 2

Norman Koren's article

Look at this image, the original is available here.

Now, let us downsample the origibal image to 256 pixels width, using Photoshops bicubic:

What we get is an image with little correlation with the original image. We got a lot of artifacts caused by aliasing. 

Next we apply "gaussian blur with radius 1" to the the original image and downsample again:

We have still some artifacts, but the image is clearly similar to the original one. The outer rings are a grey mass, but they would not be possible to resolve anyway, as four rings would fit into a single pixel.

What this demonstrates is that even good downscaling methods will introduce artifacts if the original image has high contrast details. The example also illustrates that these artifacts can be greatly reduced by preblurring the image. This is exactly what an OLP filter would do.

Now, let us look a photographic image of a test chart. The test chart used here was developed by Norman Koren, the man behind Imatest. The chart shows different bar patterns with increasing frequency from left to right. Increasing frquency is essentuially the same as finer detail. The image below is shown at 2:1 image size for easier viewing

What we see here is quite interesting. Somewhere around 90 lp/mm we start to get color artifacts, well known as moiré. At around 130lp/mm we get a grey mass with no information. to the right of the grey area we get a new line pattern. That line pattern is not real, it is a low frequency alias of the high frequency detail the sensor cannot resolve. With another word it is fake detail. Note also the color artifacts on the text below the image.

The SLT 55 has a relatively weak OLP filter as it seems. Most cameras have a bit stronger OLP filtering that supress moiré and fake detail. There are some cameras that do not have an OLP filter. These are essentially MFD cameras (except ZD that had OLP as an option), cameras with Foevon sensors, Leica M och Leica DMR, Kodak 14N/C and Nikon D800E. It is not easy to find good examples from these cameras. The Leica M9 has pretty large pixels and no OLP filter. The combination of sharp lens and large pixels make the camera prone to aliasing.

The testchart below was downloaded from Imaging Resource and developed in ACR:

 The image shows asn oscillating line pattern which is not in the original testchart. A Nikon D800 which has a higher resolution and OLP filtering does not show this kind of artifacts.

Now, that Nikon image is quite soft. We can sharpen it quite a bit:

Excessive sharpening on the Nikon will also cause aliasing:

What we can learn from this is that high microcontrast and large pixels is a combination that will lead to aliasing. DSLRs used to have OLP filtering and are normally quite unproblematic. 

My own camera, a Sony Alpha 99 SLT is far from imune to aliasing, see below:

Stopping down to say f/11 or f/16 can reduce risk for visible aliasing, as diffraction acts as a kind of OLP filter. Test charts are much more prone to show aliasing effects than everyday subjects, that doesn't say we would not have aliasing on everyday subjects, but test charts make it more obvious.

The image below is from the same Imaging Resource shots, processed by LR4.4 and using Narrow Edge preset (recommended for lamdscape). Leica M9 is on top and Sony Alpha 99 at bottom.

The Leica shows a lot of color fringing while the Sony has little, this indicates that the OLP filter on the Sony is good at supressing color artifacts.

In the monochrome version some aliasing can be seen on both camera. It is specially noticeable on the M) above the '24' mark.

Why did I not use image from the Leica M? No test shots yet. Why don't I compare Nikon D800 and Nikon D800E? Because the test shots on Imaging Resource outresolving the test target on the Nikon images! But I will add a comparison based on a previous article I have written that is here.

Real world images

The images here are used by the kind permission of Lloyd Chambers (Full size image here

The image below is from Imaging Resource, converted from raw by me. The pattern on the label is fake. This image was taken with a Pentax 645D.

Some more examples are shown on this page:

A final note

The examples shown here were choosen not to discredit any vendor, but most cameras are little prone to aliasing. To demonstrate the effect we need pictures with regular and fine detail, shot with sensors having relatively large pixels and no OLP filtering. In addition the lens needs to outresolve the sensor. 

Last Updated on Monday, 13 May 2013 21:05