Introduction

Recently, in the discussions at the large format photography site (www.largeformatphotography.info) there was a brief discussion of whether there was any advantage to be gained from scanning B&W negatives in color, and then converting to grayscale in Photoshop.

I had a suspicion that there might be a significant advantage, so I did a few little experiments.  Here are the results.

Lacking a better test target, I decided to scan my handy Stouffer 31 step density step wedge, which is contact printed onto very fine grained film.  It's not a resolution test target, but I didn't have one handy.

Scanning was done on my Microtek ArtixScan 1800f, a fairly high quality flatbed scanner popular with large format photographers who print digitally.

B&W and Color scans

This is the scan, done at 1800 DPI, 8x multisampling, 16 bit grayscale

This is the same section of the target, scanned at 1800 DPI, 8x multisampling, 48 bit color

Looking at the grayscale scan, it looks pretty normal.

Looking at the color scan, though, I immediately thought "Aha! The game's afoot!"  In particular, notice the color fringing - there's both some very pronounced green fringing, and some less pronounced red fringing.  I don't know what's causing this, although I suspect it's either an alignment issue with the scanner sensor, or else it's some chromatic aberration in the scanner optics.  Either way, it's distressing, because it's clearly a Bad Thing.

Enter the Channel Mixer

The good news is that we can do something about that fringing.  We don't need the color - we're going to throw it away anyhow.  So, the question becomes "What color, exactly, do we want to throw away, and what do we want to keep?"

Check out the following, all generated by starting with the color scan and throwing away all but a single channel using Photoshop's Channel Mixer:

Red Channel

Green Channel

And finally, the Blue Channel

And, here, we find out that it's definite - all channels might be created equal, but some are definitely more equal than others.  In this case, that green channel is quite definitely LESS SHARP than either the red channel or the blue channel.  If you look even more closely, it turns out there's a weirdness in the red channel result - look at the lower portion of the copyright symbol, where you can see a pronounced darkening at the black/white transition.  Weird, huh?  I think so, too.

So, I've decided that blue channel looks mighty good, compared to the red and green channel version.  How does it compare to the grayscale scan?

Grayscale scan

Blue channel scan

I see this as a pretty pronounced improvement.  Note that we're now imaging little bits of stuff in the white ring surrounding the C in the copyright symbol, and in the digits of the date.

How about noise?

Sharpness isn't everything in terms of scan quality.  There's also noise.

Grayscale scan

Red scan

Green scan

Blue scan

 

Now, you have to ignore the difference in tonality.  I could adjust them out, but I'm not interested in the actual grey tones rendered, because they're adjustable.  Instead, look at the texture of the tones - the 'grain' of the scan.  All of these scans were done with 8x multi-sample - in fact, they're all just differently processed versions of the same color scan (except for the topmost grayscale scan, of course).

Again, the green scan looks smoother, but that's because the fine detail of the film grain is being mushed up.  We can mush up grain whenever we want, by using the blur function in Photoshop.  What we want is to get as much detail from the film as we can.  To my eye, that's the blue scan, followed closely by the red scan.  Why?  The red scan seems to have some weird patterns in it that make it seem more noisy. The green version and the grayscale version fall a distant second; they're all mushy.

There's probably more here to be understood.  The appearance of the dark edges in the red and blue channels has me wondering what's up with those scans.  But for now, I'm gonna scan stuff in color, and use the old channel mixer to convert to grayscale.

If you do similar tests on your scanner, I'd be interested to hear what results you get.  I strongly suspect this is dependent at least on scanner model, and perhaps on unit to unit variation.