There is more to it than looking at pictures

[nma]

There have been many discussions advocating and rebutting the idea of removing the aa filter. This post discusses the basis for aa filters and refers to digital signal processing. My goal is to have a courteous discussion about the design of digital cameras. The first thing I point out is that many devices we use today are based on digital signal processing: phones, TV's, CD and other digital music.  All of these things are based on digital signal theory. The goal of digital signal processing is to represent a continuous signal with a finite number of samples. The basis for this is the Shannon sampling theorem which guarantees that if certain conditions are met that a continuous signal can be recovered from its samples. For our discussion, the samples are the data from the raw image. With the Bayer sensor there are many complications, including the rgb sensor elements which do not have the same sampling. Nevertheless, the point I want to make is that in 2010 nobody builds a sensor without having a very good idea of how it will perform. It is possible and desirable to simulate the whole process, including the aa filter, to predict the capability of the system.  The endpoint of the simulation may be a Modulation transfer function, or even a test image. The effect of the different noise components can be included in the simulation. I don't know if Canon or Nikon does this sort of thing but I would shocked to find out that they don't. This is how instruments are designed today.  If a better system could be built without an aa filter, canon and nikon would certainly do it.

[BernardLanguillier]

If a better system could be built without an aa filter, canon and nikon would certainly do it.

Agreed, but the thing is that they don't speak about it in the open, as opposed to the back makers who have been very vocal about the supposed advantages of their technological choices. So most of the discussions we hear on these forums are based on input from one side of the coin only.

Another thing is that "better" really depends on the point of view.

One of the key priorities for both Nikon and Canon (Nikon more so that Canon for some reason) has been to deliver a seamless digital experience for the photographers coming from film. That means images devoid of any digital artifact. An artifact being something that didn't exist in the real world. I believe that reduction of moire has been very high on the list of priorities among DSLR manufacturers and that alone may have taken them towards AA filter based cameras.

Cheers,
Bernard

[ErikKaffehr]

Hi,

One aspect that comes to my mind is that it is not only the image, it's also what we are doing with it?

The way we naturally analyze images is most often at actual pixels view on screen. I admit that sometimes I even zoom in, just to see better. In real life pictures are seldom used this way. The two ways pictures are presented are:

- on the web
- in print

Both methods of display include aggressive pixel manipulation. Image scale reduction does introduce aliasing phenomena, as discussed extensively in some other topics on this forum . When printing we do some pretty aggressive sharpening for output and possibly also rescale the image. The image will then be converted to some kind high frequency pattern using dithering algorithms by the printer driver.

Best regards
Erik

There have been many discussions advocating and rebutting the idea of removing the aa filter. This post discusses the basis for aa filters and refers to digital signal processing. My goal is to have a courteous discussion about the design of digital cameras. The first thing I point out is that many devices we use today are based on digital signal processing: phones, TV's, CD and other digital music.  All of these things are based on digital signal theory. The goal of digital signal processing is to represent a continuous signal with a finite number of samples. The basis for this is the Shannon sampling theorem which guarantees that if certain conditions are met that a continuous signal can be recovered from its samples. For our discussion, the samples are the data from the raw image. With the Bayer sensor there are many complications, including the rgb sensor elements which do not have the same sampling. Nevertheless, the point I want to make is that in 2010 nobody builds a sensor without having a very good idea of how it will perform. It is possible and desirable to simulate the whole process, including the aa filter, to predict the capability of the system.  The endpoint of the system maybe a Modulation transfer system, or even a test image. The effect of the different noise components can be included in the simulation. I don't know if Canon or Nikon does this sort of thing but I would shocked to find out that they don't. This is how instruments are designed today.  If a better system could be built without an aa filter, canon and nikon would certainly do it.

[nma]

Agreed, but the thing is that they don't speak about it in the open, as opposed to the back makers who have been very vocal about the supposed advantages of their technological choices. So most of the discussions we hear on these forums are based on input from one side of the coin only.

Another thing is that "better" really depends on the point of view.

One of the key priorities for both Nikon and Canon (Nikon more so that Canon for some reason) has been to deliver a seamless digital experience for the photographers coming from film. That means images devoid of any digital artifact. An artifact being something that didn't exist in the real world. I believe that reduction of moire has been very high on the list of priorities among DSLR manufacturers and that alone may have taken them towards AA filter based cameras.

Cheers,
Bernard

"
There is a lot of merit in your comment. E.g. "Another thing is that "better" really depends on the point of view." Those of us old enough to remember when CD's began to displace LP's may strongly agree. Those with golden ears believed that analog recordings, reproduced with a stylus and turntable, were more faithful to the original sound than were CD's. Early on they were certainly correct, when the sampling of the digital recording was too low. There were cases of aliasing and other unwanted sounds. But later when highly oversampled recordings were made that measured as high fidelity the argument persisted that the golden ears thought that analog recordings sounded better. There was the transister sound, thought to be harsh compared to tube amplification. Thus argued, despite the higher distortion in the tube amplification system. Does any of this sound familiar?

It could be that those who like their MFDB without an aa filter are afficited with the same sort of bias. But who cares? If they like it that way, I am not telling them they are wrong, it a point of view.  I am just pointing out that engineers generally rely on tested theory as a basis for their designs.  Tweaking by committee probably comes pretty late in the game, long after the fundamental decisions .

[nma]

Hi,

One aspect that comes to my mind is that it is not only the image, it's also what we are doing with it?

The way we naturally analyze images is most often at actual pixels view on screen. I admit that sometimes I even zoom in, just to see better. In real life pictures are seldom used this way. The two ways pictures are presented are:

- on the web
- in print

Both methods of display include aggressive pixel manipulation. Image scale reduction does introduce aliasing phenomena, as discussed extensively in some other topics on this forum . When printing we do some pretty aggressive sharpening for output and possibly also rescale the image. The image will then be converted to some kind high frequency pattern using dithering algorithms by the printer driver.

Best regards
Erik

Yes,  I find myself in agreement.  Consider a camera whose sensor was very highly sampled, say 10x more than required by the lens. The image will be totally limited by the lens.   What would we see at 100%? I claim that it would not show sharp detail at 100%.  So, as we approach that situation, e.g. with small apertures like f22, the image will not look sharp. This is not a problem. When we speak of per pixel resolution, that seems to me to be a non sequiter.