A question of print linearity... black point comp?

[xilvar]

note: I posted this on dpreview as well, but its sorta technical so I thought I might get additional responses here...

This weekend I was working on generating new quadtonerip profiles for my epson 7800. My methodology is to avoid linearization if at all possible and I was pleased to find that the 7800 was quite linear already and after some careful work I was able to make the output quite linear using only the highlight/dark/ink limit/gamma adjustments. So then I set about comparing the quadtone rip setup with the epson driver used in various ways. (ABW and color mode printing of b&w)

I use an i1 isis and sometimes a colormunki for all my color measurements. My test involved printing a 256 gray target in 5 different ways and measuring it with the isis. I was surprised to find that I was entirely unable to get the epson driver + PS to do what I understand that black point compensation is SUPPOSED to do to begin with (namely map the darkest black of the output device profile to the darkest black L=0 and then provide clean gradation from there to the white point). This puzzles me very much. Here are some charts of what happened with the 256 gray chart.



the blue lines are all the L values. You can see that the darkest black for all the techniques is at about the same place (~ L=12.5) but that all the epson driver prints had the same oddity that they clearly did NOT cleanly shade from the darkest black it ended up blocked up from RGB 0-14 instead. Anyone know why? Black point compensation was on in photoshop for all of these, and all color management for the epson driver was done by ACE inside PS.

Man, I do love how clean and smooth the quadtone rip tonal gradation is... This is my 'untoned non-neutralized' qtr profile btw so its a pleasant slightly golden color.

xilvar

[TylerB]

Xilvar, mapping from one space or profile to another using BPC merely matches the K point of one to the other in the manner you mention. What a profile, or a conversion using any intent or method, can not do is open and linearize portions of the scale that are slammed together by the nature of the device, ink, and paper combo. BPC is working with the math, and can't know that those lower L points it's working with can't be pulled apart by the device.
My suspicion is that if you were able to take a look at the profile tables indicating the lowest measurements of the device behavior, you'd find they are all blocked up. In order for a profile to correct luminosity there has to be some delineation in there in the first place for it to work with, then it can do it's best, and the various intents and BPC use would make their benefits apparent.
That's my best guess...
But of great interest in your post is the obvious superiority of using Quadtone RIP for B&W prints from UC inks in terms of photographic linearity. There are other benifits as well which are off topic, but people should be encouraged to try it out. For only $50 and the potential for greater capacity from UC inks for B&W, it really is a great development for B&W artists.
My best suggestion for Epson driver B&W work would be to go back and make tests using no profile to find driver and media settings that provides separation at every step down to 100% on your particular paper, then profile over that. Profiles are best at remapping over somewhat rational channel behavior, for lack of a better description...
Tyler

[xilvar]

Hm. That's interesting and explains the results I'm getting. I'm still puzzled as to why this has to happen though... In the case of the T9 profiles I generated them using test charts with 1728 patches of uniform color and _777_ patches of various off greys. (using a chart generation program I wrote for this prurpose). That's a LOT of samples of what exactly the printer produces for a given output rgb for the profile making program to work with. Plenty of samples that blacks are blocked up from rgb (0,0,0) through (18,18,18).

I used PM5 to generate the profiles btw. Anyway, when PM5 looks at all those samples it can easily tell what the blackpoint is and what rgb values produce all essentially the same colors super close together. So it generates a profile which is perfectly capable of generating smooth tonal greys from L=12 through L=95 when provided input RGB values from around (18,18,18) through (255,255,255). in a gamma 2.2 color space.
 
Why can't ACE just not use the blacks that are all blocked up? It has no way to unblock them. I'm not asking it to. I'm just asking it to accept that they are blocked. For example I can apply a simply photoshop transform to accomplish what I'm asking for. I can apply a levels mask with black output set to 18 and white output set to 255 just before I print. This basically accomplishes what I wanted. So I'm not clear why can't ACE do it as part of the color workflow?

So another interesting thing is that I suspect epson runs into this exact same problem with their profiles. I BELIEVE that they deal with this by discarding the measured black point and stretching the whole gamut to simply pretend the printer can print to L=0. You can sort of see that when visualizing their precanned profiles as 3d volumes. So how DO I do this myself? Can I do it with PM5? Or do I need some other profile editing tool?

xilvar