Seed dithering

[esheldon]

closes #351

Add ability to seed the subtractive dither

[esheldon]

@dstndstn does this work for you?

[beckermr]

Are there other parts of the lossy compression seed setting in the fits standard / cfitsio that we should be aware of to make sure our APIs here can support? We need not implement them OFC.

[esheldon]

Possibly. This code I added only sets the seed when you create the HDU, which seems safe. What other operations should be supported?

[beckermr]

The other options allowed by the standard are

-qt <level> - compute the seed from the checksum of the first tile of image pixels
-qN <level> where N is an integer between 1 and 10000 – use seed N (e.g., -q3010 4).

This all gets coupled into the FITS header via

FZDTHRSD - 'CLOCK', 'CHECKSUM', 1 - 10000

and here is the doc string for fits_set_dither_seed

This routine specifies the value of the offset that should be applied when calculating the random dithering when quantizing floating point iamges. A random offset should be applied to each image to avoid quantization effects when taking the difference of 2 images, or co-adding a set of images. Without this random offset, the corresponding pixel in every image will have exactly the same dithering.

offset = 0 means use the default random dithering based on system time offset = negative means randomly chose dithering based on 1st tile checksum offset = [1 - 10000] means use that particular dithering pattern

The default according to the fpackguide is

By default, the initial seed value is randomly chosen based on the
current system clock time.

So we could support the following options for dither_seed at the python level:

• 0 or 'CLOCK' or 'clock': set seed to 0
• negative number or 'CHECKSUM' or 'checksum': set seed to -1
• 1-10000: send seed through
• error for anything else

[esheldon]

yikes, I should have red the docs first

[beckermr]

A few tweaks but looking good. We'll need to adjust the tests to cover more cases and not error on -3. We should also test that -3 and -1 give the same answer.

[beckermr]

LGTM! Thanks!

[esheldon]

OK, we'll see what Dustin says