Merits of YCC processing

[Iliah]

or why I hope for better JPEGs out of D2x and D2hs

First of all, it separates luminance from colour - "Y" is luminance coordinate, "C" are two colour coordinates.

Second, YCC includes wider colour range then (s)RGB.

From the first, we can hope for much less colour artifacts and maze patterns; from the second, we can expect better colour fidelity. If new ASIC used in these cameras is converting RAW data using YCC colour representation (which would be very natural, as RAW data is actually much closer to YCC then to RGB; but conversion is more computation-intensive), we will see about 20% more resolution, lower noise, less blotches, less moire, less maze/jigsaw etc. YCC files are easier and cleaner to resize.

One of the reasons why RAWMagick interpolation is pretty clean (and slow) is because we do it in a kind of YCC colour model.

 

[Gale]

Iliah,

Thankyou for your continuing education on this.

Regards
Gale

 

[Flew]

Yes, thank you Iliah. Very informative.

Regards,

Frank

 

[Iliah]

One more thing to consider - seems that D2Hs, same as D2x, is applying white balance not in digital, but in analog. This extends dynamic range by lowering the noise and minimizing artifacts in shadows.

 

[greyflash]

Complicated but interesting Iliah. Thanks for keeping us up to date.

 

[Iliah]

I know it sounds techie. But IMHO supergigapixel chase is a bit on another side - oversimplification.

 

[Chris101]

Iliah, YCC sounds a lot like Lab. How do they differ?

Plus, I am concerned about 'analog' white balance - ie white balance being applied before the images is encoded into a RAW file. Will this mean that 'after-the-fact' white balance adjustments will be less effective and destructive?

 

[Iliah]

"Lab" is device-independent space (hence calculations to convert to Lab accurately are pretty intensive; as you need to "strip" off all device-dependent information), while YCC is a pretty simple one. What they have in common is that chroma/colour information is separated from brightness/luminance information. But YCC inherits all device-specific behavior, while Lab is free from it.

Pre-balancing is a good thing, as you use the 12 bits fully. Otherwise:
imagine Green is exposed properly, but white balancing of Red and Blue is obtained by multiplication by 2.something (typical daylight) in RAW processing. So, your Red and Blue dynamic ranges are one stop less; less then 11 bits. This "helps" to blow out highlights and to have more noise i n shadows (blotches included).

Now, if camera amplifies in analogue, this brings less noise then if it is done in digital. Comparing digital multiplication in D1 with results of analogue amplification in D2x is pretty convincing.

Second (additional) multiplication (to correct/fine tune white balance) done in RAW processing software is lossless, if done properly. Our experience with RML proved it.

 

[Chris101]

"Lab" is device-independent space (hence calculations to convert to Lab accurately are pretty intensive; as you need to "strip" off all device-dependent information), while YCC is a pretty simple one. What they have in common is that chroma/colour information is separated from brightness/luminance information. But YCC inherits all device-specific behavior, while Lab is free from it.

Thanks.

Pre-balancing is a good thing, as you use the 12 bits fully. Otherwise:
imagine Green is exposed properly, but white balancing of Red and Blue is obtained by multiplication by 2.something (typical daylight) in RAW processing. So, your Red and Blue dynamic ranges are one stop less; less then 11 bits. This "helps" to blow out highlights and to have more noise i n shadows (blotches included).

Now, if camera amplifies in analogue, this brings less noise then if it is done in digital. Comparing digital multiplication in D1 with results of analogue amplification in D2x is pretty convincing.

It has been my understanding that, for the D70, D100 and D2h cameras, the white balance setting was applied only to JPEG images in the camera, while, with NEF files, the white balance is saved in the metadata and must be applied by the converting program.

Second (additional) multiplication (to correct/fine tune white balance) done in RAW processing software is lossless, if done properly. Our experience with RML proved it.

Yes, I have also found that adjustment of the white balance with converting software (I usually use ACR) is lossless. That is, a photo shot in RAW at the wrong white balance but converted with ACR to the correct white balance is the same as one shot at the correct white balance to begin with. Here are tests I made to convince myself of this:

http://radio.weblogs.com/0101365/stories/2004/12/29/whiteBalanceRedux.html

Concerning the D2x (and maybe the D2hs?), the analog white balance amplification will be applied to the individual photo-sensors in the camera (before writing the NEF file to the card), right? So can we still expect white balance correction after that to be lossless?

 

[Iliah]

It has been my understanding that, for the D70, D100 and D2h cameras, the white balance setting was applied only to JPEG images in the camera, while, with NEF files, the white balance is saved in the metadata and must be applied by the converting program.

Indeed.

Concerning the D2x (and maybe the D2hs?), the analog white balance amplification will be applied to the individual photo-sensors in the camera (before writing the NEF file to the card), right?

To signal from individual photo sensors, right.

So can we still expect white balance correction after that to be lossless?

Given it is applied in RAW converter properly - that is without overflow in highlights and underflow in shadows. The reason why ACR is better then NC in recovering highlights; and RML is better then ACR is that NC overflows very easily; ACR - not so easily; RML - never.

 

[Chris101]

Thanks Iliah, I'm anxiously awaiting RM-Mac.