removing the AA filter

[Daniel Browning]

Hi Joe, thanks for the response.

Daniel why is the frequency of the moire in the non AA image on the rivets for the diagonal beam low?

If you're saying that you see little or no aliasing (moire) of the rivets on the diagonal beam of the non AA image, then I disagree; to me it is aliasing there as well.

If you're asking why the spatial frequency at which the aliasing of the rivets occurs is so low, I would say that it's actually not very low at all -- in fact it's quite close to Nyquist. Really bad moire occurs at much lower spatial frequencies.

if indeed each rivet is represented by more than one pixel why are some rivets missing, unless of course, these rivets are not bright chrome but colored and/or missing.

Because when you practice unsafe imaging (without a protective AA filter), you contract certain artifacts: whether or not you see the rivet depends on chance alignment of the pixel grid.

Is it not a possibility that the AA filter reconstructed missing rivets from surrounding pixels by averaging;

No. The Bayer demosaic algorithm does use a dozen or more of the surrounding pixels to interpolate color values, but that occurs at a much higher level of detail than what you're seeing in this image. Remember that this is only 4 MP -- the original is 15 MP.

are all the rivets exactly uniform?

Download the original 15 MP raw file if you want to see what they really look like.

[crames]

Because when you practice unsafe imaging (without a protective AA filter), you contract certain artifacts: whether or not you see the rivet depends on chance alignment of the pixel grid.

Daniel,

As Joe was asking, if the rivets are larger than a single pixel, as they appear to be, how could they "fall between the cracks" and disappear due to misalignment with the pixel grid?

It seems to me that some of the rivets do not appear because of subtle  differences in the direction of the glancing light. It is just as easy to pick  out areas where rivets appear in the Sigma shot but not in the Canon  shot.

Regards,
Cliff

[waynebretl]

To know for sure how much aliasing is present on the rivets, you would need two shots with the camera displaced slightly.  The form of the alias (as constructive or destructive of a small detail) does depend on the relative position of the detail and the pixel structure.  This is much more obvious in a motion imaging device if the camera moves slightly.

The first devices to invoke serious consideration of aliasing were standard definition television cameras, because of the relatively coarse scanning line structure, equivalent to 480 pixels per picture height.  This was so coarse that you could never escape a compromise between small detail contrast (vertical frequency response and sharpness) and aliasing.  In tube cameras, the aliasing was relieved by making the scanning spot blurry enough so that the scanning lines essentially overlapped their neighbors. In addition to the simple spatial aliasing, there was worse aliasing due to interlaced scanning, also relieved by the blurriness of the scanning spot. When TV cameras went from tubes to CCDs, the CCDs had to be designed to average two adjacent rows of pixels for each interlaced field in order to reduce aliasing to an acceptable level.  Then came high definition.  With about 1000 rlines (rows of pixels), the aliasing was much less objectioanble, parlty because the lens was now a prefilter for antialiasing. The CCD now could have a variable amount of averaging between adjacent rows to give a choice of more detail contrast (but with some aliasing) or less detail contrast (but cleaner).  NOTE that this is still a motion imaging case, so you have to go more towards clean than towards detail contrast, since whenever the camera moves, the variation in the alias components may be objectionably strong.

For digital still cameras, since you do not have the temporal variation in the aliasing making it so obvious, you may get away with a much higher level of aliasing, which gives greater sharpness, albeit with "false" detail.  It's a matter of taste.  My personal taste for a sensor with the high resolution of a DSLR is to keep the antialising filter and boost the detail contrast with post sharpening.  

On the other hand, in designing consumer digital converter boxes to go from high-definition broadcasts to analog TV sets, it has been found that a filter that completely eliminates aliasing looks far too soft - it's back to the same trade-offs as for a standard definition TV camera - some detail and some aliasing.

Also by the way, there is a similar principle in loudspeaker design - a mid-range driver with a small amount of distortion will make a speaker sound like it has better high frequency response than it really does - but the high frequencies are being generated in the speaker and didn't really come from the source. Depending on how much distortion is present, some sources may not sound their best.

[ErikKaffehr]

Hi,

I downloaded both images a while ago, but at that time I couldn't find the missing rivets, simply because I looked at the wrong place. Now I can see the effect clearly. The problem with this comparison is that there is no way around the fact that the D50 resolves much higher the the Sigma SD-14. We could downscale the Canon image to the same size as the SD-14.

I tried it using Photoshop "bicubic" and made the following observations:

The rivets are still there in the Canon image.

The next try I made was resize both images 200%, mostly for better viewing. The Canon image scales better. There are lots of jaggies on diagonals in the Sd-14 image while the Canon image is still pretty smooth.

So my conclusion is that the resolution advantage of the Canon is still there when the image is downscaled to 5 MPIxels. Rescaling the 5 MP images to 20 MP (200%) works better for Canon 50D. The SD-14 shows clearly more jaggies.

I don't enclose the images, as I have a problem with my Internet Provider and need to use low bandwidth solution, but anyone can retry with the original images.

Best regards
Erik

Hi Joe, thanks for the response.



If you're saying that you see little or no aliasing (moire) of the rivets on the diagonal beam of the non AA image, then I disagree; to me it is aliasing there as well.

If you're asking why the spatial frequency at which the aliasing of the rivets occurs is so low, I would say that it's actually not very low at all -- in fact it's quite close to Nyquist. Really bad moire occurs at much lower spatial frequencies.




Because when you practice unsafe imaging (without a protective AA filter), you contract certain artifacts: whether or not you see the rivet depends on chance alignment of the pixel grid.



No. The Bayer demosaic algorithm does use a dozen or more of the surrounding pixels to interpolate color values, but that occurs at a much higher level of detail than what you're seeing in this image. Remember that this is only 4 MP -- the original is 15 MP.



Download the original 15 MP raw file if you want to see what they really look like.

[crames]

The next try I made was resize both images 200%, mostly for better viewing. The Canon image scales better. There are lots of jaggies on diagonals in the Sd-14 image while the Canon image is still pretty smooth.

So my conclusion is that the resolution advantage of the Canon is still there when the image is downscaled to 5 MPIxels. Rescaling the 5 MP images to 20 MP (200%) works better for Canon 50D. The SD-14 shows clearly more jaggies.

Eric,

In contrast to the 50D file, the provided SD14 jpeg has strong sharpening halos. If you enlarge that jpeg you will also enlarge the jaggie sharpening artifacts.

I made a comparison in another thread by going back to the raw files and not sharpening either image. As expected the Canon is still better but the differences are not so glaring - or what do you think?

Cliff

[Phil Indeblanc]

Lets see some SD1 Files!!!

[BartvanderWolf]

Lets see some SD1 Files!!!

I'll show you mine if you show yours first

Cheers,
Bart

P.S. Truth in advertizing forces me to confess I don't have any