After I posted my DSLR scanning tutorial, it was suggested to me that better results might be possible if, instead of illuminating colour negatives using daylight-balanced light, that I use either a cyan/blue gel on the light source OR a similar filter (an 80A blue filter, for example) on the lens. The rationale here is that in a colour negative scan (DSLR or otherwise), the blue channel intensity is less than green, which is likewise less than red; the post-processing then corrects for this by normalizing the curves for each of the RGB channels (i.e. bringing up the blues and lowering the reds). By biasing the light source towards blue, the theory is that the blue channel intensity can be raised *in camera*, thus lowering the amount of noise in the blue channel of the scan.
Basically, by doing this, you are - in theory - "helping" the inversion process along by reducing the extent to which you have to boost the blue channel.
So, being the diligent sort, I went out and ordered a 55mm blue filter (80A) from B&H (along with some Portra 800 and Provia 100F, of course). I then shot the same 6x7 film frame with the filter either on or off my Tokina 100 f/2.8 macro. The film was an expired roll of Kodak Portra 400NC, which I shot at an E.I. of 100 in the Mamiya 7ii with the 80 f/4 lens. This roll was from a pro pack that has been stored frozen since purchase, and is known to produce good colours with a *bit* more grain than one might expect from the modern Portra 400.
One thing to note: the 80A filter has a Wratten number of 4, meaning a 2-stop loss of light. Thus, whereas I would normally image my Portra 400NC at ISO 100, f/8, 1/15 without the filter, with the filter on, the shooter speed goes down to 1/3. The ISO and aperture settings on the D800 did not change, and thus no additional shot noise is being added to the blue filtered image.
Second, these are *in no way* what I would normally present as final images. These are straight out of the scanner, as it were. No contrast curves, local contrast, colour control, etc has been performed on these images. Just the straight scan, processed by the same Photoshop action (see my tutorial link above for the links to the actions). Both images were white balanced off of their respective blank film frame spacer, as described in the main tutorial.
Anyway, on with the test.
First, the full, unfiltered image.
And a 100% crop of the above.
Now, the version exposed through an 80A filter
And the 100% crop of (approximately) the same region.
So what can we say? Frankly, theres not much in it. The 80A filter produces an image with a *slightly* bluer initial conversion (look at the slight blue cast in the road), but there's really not much there to see.
Second, the overall noise is about the same in both images (see the 100% crops). Because both crops were relatively light on blue features (where the biggest noise difference would be seen most), I decided to do a 100% crop of a region of sky. The other reason the sky should be a good place to see any noise difference (if one exists) is that the sky is generally very bright, and thus in the darkest part of the negative; this is the region that needs to be boosted the most in post-processing (that is, negative film scans are noisiest in the positive *highlights*, rather than the shadows, as with positive film or digital files).
And the 80A filter:
Exciting, ain't it?
I don't see a difference here, except (again) for the bluer tones in the sky (the non-filtered has more red in it, producing a more purple sky than the 80A filter version).
And what of the actual blue channels from each scan?
And the 80A filter:
Do you see a difference? I don't.
So what can we take away from this?
There does not seem to be a compelling reason to filter the light used to illuminate (or capture) colour negatives when DSLR scanning. Blue-filtered light produces no discernible reduction in image noise (even simply blue channel noise) relative to white light. No doubt, this has something to do with the incredible noise performance of the D800; perhaps on cameras that do not perform as well, things would be different.
But I suspect not.