Hasselblad Natural Color Solution (HNCS) - how it works (probably)

[torger]

In the past I've tried to find documentation of how Hasselblad's Phocus color model works, which they call "Hasselblad Natural Color Solution". I did not then and have not now found any description of it. If anyone knows something about it, please tell me :-).

However, now when I got my own Hassy I've poked around a bit with Phocus, and I've found out the following;

Unlike Capture One they don't use ICC profiles for camera input (unless you roll your own for repro!), and they don't use Adobe's DCP either (of course). Instead they have their own camera profile format, an XML format which you can find in the "ColorMaps" directory. The XML files are not easy to interpret though, but I've figured out some basics. (Note that the files seems to be hardcoded in the binary, so the files in the ColorMaps directory are not used by Phocus, I don't really know why they are distributed with the software)

The idea of Phocus color is that you don't need to select an input profile, the color rendition should be good for any type of light condition. In Capture One and many other raw converters you select profiles based on light condition, and in Lightroom there's the dual illuminant DCP files (they use StdA and D65 illuminants). What is Hasselblad's approach?

When playing around with my H4D-50 doing some boring test shots I've noticed that the color is indeed good in both daylight and tungsten, despite using the same profile, which is supposed to be "impossible".

However, it turns out it's indeed a multi-illuminant system, they have Tungsten (~2950K), Low Tungsten (~2100K) and Flash (~5650K), plus "Flash-Daylight" (less sure of that name, anyway same temperature as flash but daylight version), that is no less than four illuminants. Each of those illuminants have their own color matrix. Then there are color correction tables for Tungsten and Flash (there's no separate tables for the Low-Tungsten or Flash-Daylight variants, too similar I guess), each in two versions, one for the "Standard" look and one for "Reproduction", ie one subjective with more punch, and one tuned for accuracy. There's also one "neutral vector" for tungsten and flash, probably a way to describe the whitepoint with RGB multipliers.

The profile is most likely related to the working color space, which is "Hasselblad RGB" and for more recent models there's a second version which I assume is for the newer (and larger) color space "Hasselblad L * RGB". Having to care about working color space is a trace of the long history, back in the integer days you would want to have as small color space as possible to retain precision but today when everything's floating point (I hope). (As you may know Lightroom/DNG/DCP always runs in the huge ProPhotoRGB color space). The old "Hasselblad RGB" colorspace is thus quite small, most likely clipping some camera colors even, while the new is large.

The color correction tables are in some form of CbCr chroma coordinate translations, I haven't really figured that out. However it seems to me like there's no luminance coordinate, that is there's only hue adjustments which will be the same regardless of brightness. With DCP and ICC you can adjust the hue in different directions as a function of luminance if you want to (not sure how useful that is though, but I know Adobe does it in their profiles).

They don't have a profile per camera model, rather per sensor as it's that what sets the color response. This means that the color of H3DII-50/H4D-50/CFV-50 should be exactly the same as they all use the Kodak KAF-50100 sensor, as well as between CFV-50c and H5D-50c which share the Sony CMOS. I haven't checked if they have tried to match the look between different sensors though, but I have heard that they match quite good.

As far as I understand Phocus will automatically select illuminant matrix and chroma correction LUT based on the white balance you set in Phocus, and blend between them for intermediate values, that is the same way dual illuminant DCP works in Lightroom. The difference from DCP is that Phocus profiles have more illuminants, and is more sophisticated in that aspect, but have a bit simpler LUT. On the other hand will the simpler LUT make sure that they're no nasty non-linearities which you can have in (bad) DCPs.

In all I'm quite impressed with the color model in Phocus, it's simple towards the user and advanced under the surface.

edit: it's called Hasselblad Natural Color Solution, not System :-)

[Joe Towner]

Very cool read.  It's funny when you look under the hood of how Phase/Leaf/Hassy/Pentax all use the same chips, yet produce different 'ideal images'.

[Chris Livsey]

The profile is most likely related to the working color space, which is "Hasselblad RGB" and for more recent models there's a second version which I assume is for the newer (and larger) color space "Hasselblad L * RGB". The old "Hasselblad RGB" colorspace is thus quite small, most likely clipping some camera colors even, while the new is large.

They don't have a profile per camera model, rather per sensor as it's that what sets the color response. This means that the color of H3DII-50/H4D-50/CFV-50 should be exactly the same, as well as CFV-50c and H5D-50c.

At last some light is shed under that particular bonnet, many thanks.
I have noted in bold a couple of points in my quote and I would appreciate your comments on them.

1) If the "old" colour space is possibly clipping surely the colours across models will , in some circumstances, be different?

2) Do we/you know what the cut off point is? ie which models switch to the newer, wider, colour space? Or does your bracketing of models indicate that the change occurred with the H3DII onwards?
I realise with assiduous searching I could find out from the spec sheets but asking is easier 

[torger]

"Hasselblad RGB" is about the same size as AdobeRGB, while "Hasselblad L * RGB" is about the same size as ProphotoRGB.

AdobeRGB is too small to fit all colors a camera can record (and actually even ProPhoto misses a little bit of extreme blue but that's more theoretical), and same counts for Hasselblad RGB. However very few natural colors have such extreme saturation that it becomes a problem, so it's not that big of a deal. When raw converters changed their internal math from integer to floating point (because computers became fast enough) some also became "clipless", meaning that you can represent negative values and values larger than the clip point and in that case you could be using a small working color space. We had it like that in RawTherapee for a while but it became awfully messy with many of the internal algorithms. I don't know how Phocus makes the internal calculations, but I would not count on that it's clipless.

Although it will in 99% of the cases make no difference as colors rarely are that saturated I find no reason not to use Hasselblad L * RGB all the time. As said Lightroom always run in ProphotoRGB internally and don't allow the user to change the working space. Having a working space setting at all is just an indication that the software has a long history from the time when we needed to make calculations with integers to keep up speed. Note that working space has no connection to the output space which you set separately. If you export to 8 bit JPEG it's still a good idea to use as small space as possible to avoid banding, sRGB is generally enough.

When I said "more recent models" I should have said "more recent sensors". The profiles are per sensor, not per model. It seems like the Kodak 16/22/31/39 megapixel models have only a "Hasseblad RGB" profile (which is probably converted through regular color math to Hasselblad L * RGB when needed, this can be done without clipping or with clipping, I don't know how they do it), while the Kodak 40 & 50 and Dalsa 60 and Sony 50 have for both. Which profile is which is a bit foggy though, as the files in ColorMaps directory is not used directly I have to guess a little bit.

[Chris Livsey]

When I said "more recent models" I should have said "more recent sensors". The profiles are per sensor, not per model.

I should have realised that.
Did Hasselblad code this internally I wonder or was an external "house" given the job? It seems, which we probably knew already, that they had a unique approach.
How does this work when other sensors are worked in Phocus, does it just pick up on the Apple profiles as I read elsewhere?

[torger]

I should have realised that.
Did Hasselblad code this internally I wonder or was an external "house" given the job? It seems, which we probably knew already, that they had a unique approach.
How does this work when other sensors are worked in Phocus, does it just pick up on the Apple profiles as I read elsewhere?

I'm quite sure Hasselblad did this internally, the model has been there for a long time (from the first release perhaps?), they haven't just talked about it much other than in incomprehensible market speak.

Looking at naming of files they seem to have this type of profile for some Leica system, but otherwise only their own cameras. The support for Canon Nikon etc is only available on Mac OS X (I run Phocus on Windows so I haven't even tested opening other than Hasselblad files), and then only because the OS X APIs allows reading raw files very easily, but you then get a very generic type of color handling. Personally I wouldn't use Phocus for any other cameras than Hasselblad cameras.

[Paul2660]

So I guess it's safe to to say that color from CMOS and CCD look the same, at least with Hasselblad. 

Thanks for the information

Paul

[torger]

So I guess it's safe to to say that color from CMOS and CCD look the same, at least with Hasselblad.  

As said they have different profiles per sensor, while H3DII-50/CFV-50/H4D-50/H5D-50 all using the Kodak KAF-50100 sensor share the same profile, the Sony CMOS models (CFV-50c, H5D-50c) have their own profile.

I would expect a quite different look between the two (edit: post below says Hassy match different sensor colors quite well), but still I haven't seen any side-by-side testing, ie some representative scenes shots by both a H5D-50 and H5D-50c. Or wait a minute, maybe I have.... I have a faint memory of that I have seen such a test somewhere, I just don't remember where now.

[lelouarn]

So does this mean that all Hasselblad CCDs (and even CMOS) have the same Color Filter Arrays ? The only variable is the chip, not the microlenses in front of it ?
Or is the CFA's response taken care in some other way ?

[torger]

So does this mean that all Hasselblad CCDs (and even CMOS) have the same Color Filter Arrays ? The only variable is the chip, not the microlenses in front of it ?
Or is the CFA's response taken care in some other way ?

No, I probably did not describe it clear enough in the original post, I shall edit it to make it a bit clearer. There's one profile per sensor. For example H5D-50 and CFV-50 share the same profile as both models have the Kodak KAF-50100 sensor. However the new Sony CMOS model H5D-50c and CFV-50c have a different profile as they use a different sensor.

This approach is not unique, but I haven't checked how for example Phase One does, P65+ and IQ160 could share the same profile as they share the same sensor and same CFA, but sometimes one can do a little change anyway to differentiate the products. It does happen that sensors are slightly upgraded and get new CFAs, I don't really know how the history has been with the Dalsas Phase One is using.

[Dustbak]

Not sure how it is between the CMOS and CCD but HB takes great effort in getting the same color response for every type of camera.

I use a H4D60 and a H5D40 which have different sensors but the colors are really close. So close that I have no hesitation of using/combining images from both in one shoot.

I would not be surprised when the CMOS is pretty similar to the CCD.

[torger]

Not sure how it is between the CMOS and CCD but HB takes great effort in getting the same color response for every type of camera.

I use a H4D60 and a H5D40 which have different sensors but the colors are really close. So close that I have no hesitation of using/combining images from both in one shoot.

I would not be surprised when the CMOS is pretty similar to the CCD.

Thanks, that's very interesting info, I haven't really done any comparisons between different models myself. But reading this article from Hasselblad themselves:
http://press.hasselblad.com/media/1096/2010-09-22_background_hncs_en.pdf
it appears like they've devised one way to represent colors that's "best", at least according to their taste, and they probably then strive to make their cameras look that way. My guess though is that they would make a decent "global" match, but then optimize profiles for each sensor individually. It's not a good idea to work too much against the CFAs with the profile as you then can get some nasty non-linearities.

Hasselblad's own description talk only about "a single profile", but it's really not that technically correct. A multi-illuminant profile like they have is actually several profiles baked into one, with auto-choice/blending through white balance setting. What is unified though is that they don't have a particular profile for skin tones and another for products and a third for landscape. If that's good or bad I guess is a matter of taste.

[eronald]

C1 have profiles for things like products which will drag close to unsaturated colors to central gray.
This has the advantage also of dealing with mild IR contamination, sensor miscalibration etc.
The idea is always the same - give the customer the color she wants rather than the color the camera sees ...

Edmund

[Zerui]

Thank you for a fascinating insight into Hasselblad colour management.
I find that Phocus does a really excellent job with the colour from my H3DII50. And the user interface makes it easy to get the results you want.
One of my favourite demonstrations is the attached image of a swimmer in a sea coloured by reflections of sky and shore, featuring diffusion rings from the sharp skyline, and upwelling light from caustics on the seabed, refracted and dispersed at the sea surface.  The complicated physics, which has only recently been sorted out, goes under the name Natural Focusing.  The point is that the swimmer's skin tones are right, and so too are the water colours.  Quite a test!  Zerui

[Doug Peterson]

C1 have profiles for things like products which will drag close to unsaturated colors to central gray.
This has the advantage also of dealing with mild IR contamination, sensor miscalibration etc.
The idea is always the same - give the customer the color she wants rather than the color the camera sees ...

Edmund

Edmund, in the case of the newly released Capture One Cultural Heritage Edition the profiles include reproduction profiles made for absolutely accurate, non-interpretative color. These profiles are also built specifically for use with linear curves.

[deejjjaaaa]

This approach is not unique, but I haven't checked how for example Phase One does

it does greatly sometimes... once upon a time C1 (v6.*) was released with a standard profile for Pentax K5 (Sony made 16mp sensor) just copied 1:1 from profile from Pentax K7 (Samsung made 14mp sensor)... granted that was too small marketshare to bother with and not P1 own camera.

moral of the story - look inside.

[eronald]

Edmund, in the case of the newly released Capture One Cultural Heritage Edition the profiles include reproduction profiles made for absolutely accurate, non-interpretative color. These profiles are also built specifically for use with linear curves.

Sure  - who except a museum is going to complain if their photographic system magically cleans up color blemishes ? 

Edmund

[EricWHiss]

Edmund, in the case of the newly released Capture One Cultural Heritage Edition the profiles include reproduction profiles made for absolutely accurate, non-interpretative color. These profiles are also built specifically for use with linear curves.

Doug,
Thanks for bringing this new version of C1 to my attention.  Very interested in that! Will have to have a look.
Eric

[mi-fu]

It is a very interesting read, torger. Thanks for the analysis. It answered some questions in my mind for a long time 

[soren]

Anders,

I did a test with the HD5 50c and a HD3 II 50 almost a year ago. Just changed the camera on the tripod both with 28 mm. If you like to have a look I could send them over to you. Send me a pm

Regards
Sören