Those wacky engineers (2D Fourier Transforms)

[J A C S]

So here are the 2D Fourier Transforms (2D FTs) of the black frames for a camera over each of it's 16 ISO settings from upper left by row to lower right (adjusted as usual to enhance patterns):


256x256 RGGB from the center of the image.

The first ISO setting with clear signal processing has a very unusual pattern.
I've never seen this dense grid before.

And rather than staying steady or increasing with ISO setting it fades, then comes and goes.

Very frustrating because (observed) read noise has a totally bizarre pattern.

I'm starting to think this is a lens correction that can not be turned off.

Since FTs are reversible I tried the following crude test from row 2 column 4:



This concentric pattern looks like a lens correction like light falloff.

Not sure why the effect varies so much at different ISO settings; but this seems to be a start.

It is a fixed lens (not ILC) camera so they may have simply baked this in .
Regrettable since it's a DNG and they could simply put in the appropriate Opcode.

Aha! These files did in fact get past my collaborator Quality Control (QC).

Exiftool confirms that Distortion Correction was On.
PDF of the user manual says it's a settable menu option.

So my contributor for this camera didn't fully follow the protocol and I didn't catch it until now.

Did you repeat the calculations with the Distortion Correction off?

What you see in you simulation is a standard Airy pattern created by the disk(s) and also a finer periodic pattern created by the periodic one in the FT. Does not look like vignetting correction to me, and certainly not like a distortion one.

 

[sharkmelley]

Did you repeat the calculations with the Distortion Correction off?

What you see in you simulation is a standard Airy pattern created by the disk(s) and also a finer periodic pattern created by the periodic one in the FT. Does not look like vignetting correction to me, and certainly not like a distortion one.

I would have thought that a distortion correction is different to a vignetting correction. On the other hand I can't imagine how a distortion correction might be applied to a raw file.

Mark

 

[Joe Pineapples II]

Did you repeat the calculations with the Distortion Correction off?

What you see in you simulation is a standard Airy pattern created by the disk(s) and also a finer periodic pattern created by the periodic one in the FT. Does not look like vignetting correction to me, and certainly not like a distortion one.

I agree - distortion correction is a non-uniform, anisotropic stretching of the image that is much larger in the corners than the centre. If you input a pure sinusoid and applied distortion correction it would potentially shift vertical and horizontal frequencies (by stretching them), and smear them out a bit (because the stretch is non-uniform), but I can’t see how it would produce the pattern you showed in the FT. If your samples are taken from the centre of the frame, it wouldn’t have much effect at all, unless the lens has massive distortion.

J.

 

[bclaff]

Did you repeat the calculations with the Distortion Correction off?

What you see in you simulation is a standard Airy pattern created by the disk(s) and also a finer periodic pattern created by the periodic one in the FT. Does not look like vignetting correction to me, and certainly not like a distortion one.

I agree - distortion correction is a non-uniform, anisotropic stretching of the image that is much larger in the corners than the centre. If you input a pure sinusoid and applied distortion correction it would potentially shift vertical and horizontal frequencies (by stretching them), and smear them out a bit (because the stretch is non-uniform), but I can’t see how it would produce the pattern you showed in the FT. If your samples are taken from the centre of the frame, it wouldn’t have much effect at all, unless the lens has massive distortion.

Here's a gross example:


Left crude distortion correction; middle FT of crude; right actual 2D FT from the camera

--
Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

 

[Joe Pineapples II]

Did you repeat the calculations with the Distortion Correction off?

What you see in you simulation is a standard Airy pattern created by the disk(s) and also a finer periodic pattern created by the periodic one in the FT. Does not look like vignetting correction to me, and certainly not like a distortion one.

I agree - distortion correction is a non-uniform, anisotropic stretching of the image that is much larger in the corners than the centre. If you input a pure sinusoid and applied distortion correction it would potentially shift vertical and horizontal frequencies (by stretching them), and smear them out a bit (because the stretch is non-uniform), but I can’t see how it would produce the pattern you showed in the FT. If your samples are taken from the centre of the frame, it wouldn’t have much effect at all, unless the lens has massive distortion.

Here's a gross example:

View attachment 5843569
Left crude distortion correction; middle FT of crude; right actual 2D FT from the camera

--
Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

Very likely I'm missing the point, as I make no claims to know about digital image processing; but I'm not seeing why concentric black and white rings relates to distortion correction.

As a quick back-of-the-envelope, here is the FT of some 1-D white noise:



If I apply a crude moving average filter (length 5 with weights [0.2, 0.2, 0.2, 0.2, 0.2] in this case - chosen quite abitrarily) to the input noise before the FFT, I get a low-pass filter with the kind of sidelobes you see in some of the images:





I recognise that this is not exactly what you are seeing, but my first thought was that type of structure is characteristic of applying a window with a fairly sharp cut-off. Obviously a moving-average filter with all coefficients equal is an extreme example of that.

Like I say - I'm no image processing expert, so it is quite possible I'm entirely missing the point of your original post.

J.

 

[bclaff]

Did you repeat the calculations with the Distortion Correction off?

What you see in you simulation is a standard Airy pattern created by the disk(s) and also a finer periodic pattern created by the periodic one in the FT. Does not look like vignetting correction to me, and certainly not like a distortion one.

I agree - distortion correction is a non-uniform, anisotropic stretching of the image that is much larger in the corners than the centre. If you input a pure sinusoid and applied distortion correction it would potentially shift vertical and horizontal frequencies (by stretching them), and smear them out a bit (because the stretch is non-uniform), but I can’t see how it would produce the pattern you showed in the FT. If your samples are taken from the centre of the frame, it wouldn’t have much effect at all, unless the lens has massive distortion.

Here's a gross example:


Left crude distortion correction; middle FT of crude; right actual 2D FT from the camera

--
Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

Very likely I'm missing the point, as I make no claims to know about digital image processing; but I'm not seeing why concentric black and white rings relates to distortion correction.

The image "distortion correction" in this camera appears to be dominated by a light falloff correction which would be characterized roughly by concentric circles.

...

Like I say - I'm no image processing expert, so it is quite possible I'm entirely missing the point of your original post.

At the time of the original post I didn't understand what was going on; there was no "point" it was a query for ideas.

--

Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

 

[J A C S]

Here's a gross example:


Left crude distortion correction; middle FT of crude; right actual 2D FT from the camera

--
Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

Very likely I'm missing the point, as I make no claims to know about digital image processing; but I'm not seeing why concentric black and white rings relates to distortion correction.

The image "distortion correction" in this camera appears to be dominated by a light falloff correction which would be characterized roughly by concentric circles.

That would be a very unusual vignetting correction. The conventional one would just serve as a high pass filter for the FT and make the noise more hight frequency one, roughly speaking. It may create some additional minor effects due to the "wrapping".

 

[Joe Pineapples II]

The image "distortion correction" in this camera appears to be dominated by a light falloff correction which would be characterized roughly by concentric circles.

[...]

Vignette correction would look more like this, wouldn't it?



J.

 

[bclaff]

The image "distortion correction" in this camera appears to be dominated by a light falloff correction which would be characterized roughly by concentric circles.

[...]

Vignette correction would look more like this, wouldn't it?

Not that smooth. Remember, the values get quantized so there will be concentric "peaks" . That's what my crude rings do.

--
Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

 

[J A C S]

The image "distortion correction" in this camera appears to be dominated by a light falloff correction which would be characterized roughly by concentric circles.

[...]

Vignette correction would look more like this, wouldn't it?

Not that smooth. Remember, the values get quantized so there will be concentric "peaks" . That's what my crude rings do.

There would not be concentric peaks but concentric circles where the amplitude jumps (or its derivatives do).

BTW, why don't you FT a 256x256 patch away from the center to end all speculations?

 

[Joe Pineapples II]

The image "distortion correction" in this camera appears to be dominated by a light falloff correction which would be characterized roughly by concentric circles.

[...]

Vignette correction would look more like this, wouldn't it?

Not that smooth. Remember, the values get quantized so there will be concentric "peaks" . That's what my crude rings do.

4 bits




8 bits


12 bits

It's still pretty darned smooth!

J.

 

[bclaff]

That was a Pentax MX-1.

And I now have another Pentax camera, the KP, with equally baffling results.

Aargh :-x