On the use of dithering when decompressing lossless NEF files

[mallman]

Hi.

I was looking into what uncorrectedRawValues does when set to false (the default), specifically for D850 14-bit lossless compressed NEF files. Because these are "losslessly" compressed, the decompression tone curve is the 14-bit identity vector.

I understand the usefulness of using dithering when decompressing using a non-trivial decompression curve. However, in this case it seems that the dithering is just introducing some noise into what should be a identity transform.

Is this a bug? Should Rawspeed bypass dithering (or tone curve decompression entirely) when working with an identity tone curve?

This could be a performance improvement as well. Just do the lossless Huffman decompression and skip the tone curve decompression.

[LebedevRI]

I understand the usefulness of using dithering when decompressing using a non-trivial decompression curve.

Related, i really need to look into at least the dithering-during-black/white-scaling code,
and finally unify their behavior and get rid of the SSE2 codepath.

However, in this case it seems that the dithering is just introducing some noise into what should be a identity transform.

Is this a bug? Should Rawspeed bypass dithering (or tone curve decompression entirely) when working with an identity tone curve?

I'm not familiar with that particular piece of code.
It does seem like it should be a no-op for identity curve.
I'll take a look.