Profiling "raw" scans of negative film

[bernard_lazareff]

<p><em><strong>Goal</strong></em>: obtain true colors (or a good first approximation to reduce fiddling) from scanned negative film, using a shot (same film) of an IT8 reflective target. I use vuescan and a Nikon LS-2000.<br>

<em><strong>Previous attempts</strong></em>. Profile film inside vuescan. More detailed description: preview blank frame, lock exposure, preview again, lock film base color, scan taget (color balance: none), profile film, save and activate profile. Scan again the IT8 (color balance:neutral); the lighter patches of the gray scale (GS) of the IT8 have a definite blue cast(???). Neutralize that with a right click; now in color balance:manual. Start scanning regular frames. Bluish cast for the first dozen frames, then (without reason) shifting into a yellow cast(???). Pull hair in despair.<br>

<em><strong>Current effort</strong></em>. Leave a minimum of initiative to vuescan. Again, lock exposure and film base color; color balance: none; generic color neg; output color space: deviceRGB; output: 48bit Tiff; no tiff profile. I call this a "raw" scan for lack of better word, fully aware it's not the same as a vuescan raw. Vuescan does the conversion to positive, and nothing else. More important: everything inside the vuescan black box is frozen (at least I believe).<br>

Scan IT8 target. Produce profile "X" with Argyll (kudos to the developer of Argyll!). Profile type: shaper-matrix. Scan regular frames. Using Picture Window Pro (probably same can be done with PS) for each "raw" scan <em>assign</em> the "X" profile, then <em>convert</em> to AdobeRGB. IT8 target looks OK. So do the scene shots, <em>mostly</em>.<br>

<em><strong>Current problem</strong></em>. Looks like the white patch of the IT8 target is mapped (by the profiling) to maximum white in the colorspace. Fair enough. But, whenever a part of the scene (clouds...) is recorded on film at a higher point (on the HD curve) it saturates.<br>

<em><strong>Possible solutions; advice solicited</strong></em>.</p>

<ol>

<li>Modify the X, Y, Z values (and/or L, A, B) in the IT8 target calibration data, to correspond to a light level 1/4 or 1/8 of original; should leave some headroom for lighter tones.</li>

<li>(to be combined with [1] above): change from shaper-matrix to gamma-matrix profile, because I expect the polynomials in the shaper to behave poorly when extrapolating.</li>

<li>Generate synthetic IT8 target combining data from two different (known) exposures, at, e.g., nominal and nominal +3 stops. Need to fudge values as in [1] above</li>

<li>Discover magic option in profile software, to the effect that X=Y=Z=100 maps to white hue, but (say) 1/8 of max luminosity.</li>

</ol>

<p>Your input is welcome. I am not interested in solutions involving "auto white balance" and the like; no piece of software can guess what the scene really was like.</p>

 

[bernard_lazareff]

<p>@ Les Sarile<br>

I started with Nikon Scan. My recollection (many years ago) is that it was either auto-colors of some kind, or fiddle with sliders until image "looks right". Vuescan has (in my eyes) one important plus: scan from preview, that saves significant time. If you are familiar with Nikon Scan, can you please tell me (saves re-installing, etc) if it allows:<br>

- to avoid scanning twice (preview + actual scan)<br>

- to accept any valid film profile created externally (Argyll or other)<br>

Thank you for your interest and feedback</p>

[bernard_lazareff]

<p>Self-solution. <br>

As often happens, writing down my problem helped me solve it. The magic bullet is the -U option for argyll's colprof command, that provides just what I needed, some headroom in the color space above teh white patch of the IT8 target. Specifically I used:<br>

colprof" -v -D"RealaLS2kC" -qm -ag -U4.0 Mire-02-1<br>

where Mire-02-1.ti3 is the output of the scanin command (reads patch values). <br>

Profiling my "raw" scans (see in OP meaning of "raw") with the resultant profile results in tif files that are intrinsically "dark", because thay now have headroom above the IT8 white. Possibly -U4.0 was overkill, but at least I'm safe in the sense that high values are protected. Now I can set the black and white points to my liking: all the information is there. <br>

First example: image produced <em>witout</em> the -U4.0 option. Cloud highlights are blown. </p><div></div>

[bernard_lazareff]

<p>Next, same with -U4.0, then set white (and black) points at edges of histogram</p><div></div>

[bernard_lazareff]

<p>Another example, first using profile created without -U option</p>

[bernard_lazareff]

<p>Another example, first using profile created without -U option<br>

(repeat, made mistake trying to upload enormous tiff file)</p>

[bernard_lazareff]

<p>Another example, first using profile created without -U option<br>

(repeat, made mistake trying to upload enormous tiff file)</p><div></div>

[Lou_Meluso]

<blockquote>

<p>using a shot (same film) of an IT8 reflective target.</p>

</blockquote>

<p>Being a piece of photographic paper, it's not an ideal target to shoot with film. Try an Xrite Passport color checker. Use standard, clear sky, noon direct daylight for your centerline. Bracket your exposures in 1/3 stops to find the ISO/ASA your scanner likes best with nominal settings. Customize from there.</p>

[bernard_lazareff]

<p>And the second example, using the profile created with -U4.0</p><div></div>

[bernard_lazareff]

<p>@ Louis Meluso</p>

<blockquote>

<p>being a piece of photographic paper, it's not an ideal target to shoot with film</p>

</blockquote>

<p>Could you please explain what specifically is not-ideal? The only problem I can think of is specular reflections, or, said differently, the light beam geometry of taking a picture under "natural" conditions differs from the (a?) definition of reflection density and PCS coordinates. In what way would a ColorChecker avoid that issue? </p>

<blockquote>

<p>Bracket your exposures in 1/3 stops to find the ISO/ASA your scanner likes best with nominal settings</p>

</blockquote>

<p>One beauty of color negative film is the exposure latitude. I'd rather not put myself in a situation where exposure is critical to 1/3 stop. Possibly I miss your point. <br>

Anyway, thank you for your interest and response. </p>

[bernard_lazareff]

<p>@ Les Sarile</p>

<blockquote>

<p>Sample below shows the Coolscan with Kodak Portra 160VC using Vuescan's specific film profile (...) straight-up auto scan using Nikonscan</p>

</blockquote>

<p>I definitely won't champion the presets of vuescan vs any other software. Indeed, I wrote in my OP: Leave a minimum of initiative to vuescan. <br>

This said, the NikonScan Neutral looks better than <em>most</em> vuescan presets (I gave up on those), but... was the dress really desaturated red as in the NikonScan thumbnail, or brighter as in Vuescan Autolevels? Or...? As I already stated, how can a piece of code decide without a calibration reference, just from data internal to <em>one</em> picture, decide what were the true colors? That is the reason why I prefer to use profiling. </p>

<blockquote>

<p>BTW, how many frames have you already scanned</p>

</blockquote>

<p>Definitely less than 20k. And not trying to establish a position as top scanner dog. Just to share some specific experience potentially useful to others. Beyond that, to each his/her own. </p>

<blockquote>

<p>what kinds/brands of films are you scanning</p>

</blockquote>

<p>Mostly Reala, some Portra, Fujicolor200, Kodak UC200. </p>

 

[Uhooru]

<p>Nikon Scan does have a preview, and it will give you a view before adjustments and a tab with a view after adjustments. </p>

[Lou_Meluso]

<p>Perhaps not ideal was a poor choice of words. I <em>prefer </em>using a paint reference as opposed to an Ektacolor print for it's matte surface, dimensional stability and over all long term color stability. I know they are carefully made but it is still just a color print. You expressed some color shift problems in your results. The standard used, and it's condition, can be a factor. Not everyone has a digital color checker SG hanging around but the Passport is more commonly found. The color, consistency and geometry of the source illuminant, which you did not mention in your OP, can also be a source of profiling variability. That is why I mentioned it.<br>

As to exposure, the idea is not to bracket every exposure by 1/3 a stop, only the standard image that will be used for further calibration. Scanners vary somewhat on what is interpreted by the device as "normal" negative density. It's scanner/emulsion/lab specific. <br>

These are not hard and fast answers to your problem, Bernard, just some ideas to help you problem solve. Apparently you have now found some solutions you are happy with...good.</p>

[wogears]

<p>Have you tried ColorPerfect? I get excellent results with a minimum of fiddling.</p>

[bernard_lazareff]

<p>@ Les. I had a look at ColorPerfect literature; I can see that their concerns meet mine in several respects: leave a minimum of initiative to the scanning software, preserve highlights, use film-specific data. I trust it is a valid solution, but I can't evaluate all possibilities, just need to find something that works for me.<br>

@ Louis. I can see the benefit of a smaller size color reference card. Indeed, carrying the Wolf Faust A4 target on location is not very convenient. I might give a try to a ColorChecker or ColorPassport. As to stability, after 10 years of using the Wolf Faust target, I bought a new one: putting the two side by side my eyes cannot see any difference; should be good enough to profile photos, that are generally evaluated relying on visual memory, without the original as reference.</p>

[Robin Smith]

<p>The main difference I see in Bernard's shots are in brightness (sensor exposure) and/or contrast: the color balance seems to be very similar if not the same on my screen.</p>

[bernard_lazareff]

<p>@ Robin Smith<br>

Both pics were processed in a similar way: use vuescan to produce files with minimal and identical processing: same exposure, same film base color; Both were processed with <em>almost</em> the same profile, except in the second case I arranged for the profile to reserve some luminance headroom above the white patch of the IT8 chart. For the second pic, the only post-processing was to set the black and white points (the histogram had a lot of headromm above highlights). The first one clips the highlights, and as a consequences, its midtones are placed higher than in the second pic.<br>

The colors (hues) had better be the same.The images I showed are not meant to be in final form; probably an S-curve would be appropriate. BUT, that intepretive part had better start from technically accurate colors.<br>

Below, the histograms of the two pics.<br>

To summarize.</p>

<ol>

<li>Profiling allows to obtain accurate colors without fiddling or guesswork (whether from the operator or some black-box recipe of the scanning software).</li>

<li>Plain vanilla profiling clips at the luminance of the IT8 chart (more precisely at the luminance of a fictitious perfect white reflecting target)</li>

<li>The -U option of argyll/colprof allows to reserve some headroom and to <em>decide</em> where the white (and black) point should be placed. The 4.0 value I used in initial test is probably overkill. </li>

</ol><div></div>