Drizzle generates a matrix

[Sebastiano]

Hello,

I am lately experimenting a bit with DrizzleIntegration and I noticed a sort of matrix in my drizzle integration file.

Am I doing something wrong? Is there a way to avoid this matrix?

Files added: 79 x 900s subs total, 42 L + 37 RGB.

Dithering was 4 to 6 pixel wide.

[jkmorse]

I have never seen this.  Are you sure this is only at the drizzle integration stage and is not in ImageIntegration as well?  Also, are you using the default 2/90 settings in Drizzle?

[Sebastiano]

Hello Jim,

Yes, it only happens after the DrizzleIntegration and not in the ImageIntegration. I used default drizzle settings.

[Sebastiano]

I have just re-aligned and averaged 22 luminance frames, then I did a Drizzle integration.

This is how the matrix appears.

[Sebastiano]

I've just found this has to do with the Drop Shrink option.

In the last drizzle_weight screeshot I used a Drop Shrink value of 0.90.

When increased to 1.0 the result improves, but still it's not enough to totally remove that matrix.

[Juan Conejero]

Dithering was 4 to 6 pixel wide.

Apparently your dithering is too accurate. For drizzle, dithering must be performed by fractional pixels, ideally following a random pattern to ensure uniform pixel coverage. If the images are always displaced by integer pixels (or nearly), drizzle will tend to generate aliasing artifacts.

[Sebastiano]

Thank you Juan for your explanation.

Could it also be the SuperBias involved into this?

Maybe such median algorithm could have caused the pattern, but it's not visible in the ImageIntegration file though.


While testing, look how weird a drizzle_weights image appeared:

[vicent_peris]

Hi,

I would load all the image set into Blink to make a visual inspection of all the image. Maybe some of them have something strange...?


V.

[Sebastiano]

Hello Vicent,

I did see them into Blink and apart some lightness differences between frames and dithering movements I can't see anything strange.

About the too accurate dithering, all frames have been acquired during six different nights with a portable setup, everytime positioning the rotator position manually with some arcminutes of rotation error, so I wouldn't bet it was so accurate.

[vicent_peris]

Please could you upload the data set?

V.

[Sebastiano]

Yes, I am going to.

It will take some time though.

[whwang]

A wild guess is that one (or few) of the images have much higher background than the other images.  If images are not normalized for background brightness, drizzle will produce something like this.  Did you turn on the "Normalization" function in the ImageIntegration tool? 

My question to Juan is whether the normalization is carried from ImageIntegration to DrizzleIntegration.  It is not obvious to me since there is not an "Enable image normalization" option in DrizzleIntegration, but there are options like "Enable pixel rejection" and "Enable image weighting" etc.

Cheers,
Wei-Hao

[Sebastiano]

Hello Wei-Hao,

Your guess could be right because some of the luminance frames have been acquired still under a good sky but that night the sky was brighter because of humidity and this has produced higher backgroud adu.

Normalization in ImageIntegration was set to Scale + Zero offset.

Your question is interesting, because DrizzleIntegration might have not performed that then.

Let's wait the response of Juan.

[Juan Conejero]

Normalization parameters are always stored in .drz files by ImageIntegration, and are always applied by DrizzleIntegration. These parameters are the statistical scale and location estimates for each input image, which are computed with the IKSS algorithm by default (see a description of this algorithm in the Image Normalization section of ImageIntegration documentation).

Note that the normalization data that drizzle uses is the output normalization of ImageIntegration, that is, the parameters used for pixel combination, not the rejection normalization parameters. The output normalization is "additive with scaling" by default. All this stuff is described in the documentation.

It's true that there are no options to disable image normalization in DrizzleIntegration, as doing so doesn't make any sense IMO. We could add an option to disable it, though, if there are good reasons.

Let us know when you have uploaded the data set, since I am also interested in this test case.

[whwang]

Hi Juan,

Thanks for clarification.  As you said, it makes no sense not to normalize the images in drizzle.  I just wonder about this since there is no obvious indication whether the normalization is applied or not.  It's good to know that it is.  Now the problem Sebastiano encountered must be caused something else.

BTW, I posted a question in the Bayer drizzle thread.  Do you mind taking a look?

Cheers,
Wei-Hao