Sepia Blues

Almost two years ago I wrote an entry about in-camera sepia, wondering if in fact a sepia transofrmation could provide a photo with more tonalities than a tyical 8-bit black and white.

At the time, I assumed that the crucial yes or no part of that answer would involve the B&W conversion -- if it was done before the sepia rotation of the color space, or after -- that is, if the # of B&W values was fixed to 256 and then rotated, or if it was in a higher precision RAW format, rotated, and then quantized to eight bit. In the latter case, you'd have more tones.

I was half right, and it was the poor half. Looking closely at LX2 sepia images like the one above, it's become clear that the B&W is converted and quantized first. And the part I was wrong about I didn't expect -- that the LX2 crates sepia images not through a matrix transform in RGB color space, but simply by adding a color to the B&W image. The color appears to be (+30R,-18G,-38B). That is, raise red, and drop green and blue, by a simple addition.

Since the values are already clipped to 8-bit, this means a loss of highlights in the red channel and a loss of shadow detail for green and blue. Since all three are made from the same B&W original, it's possible to rebuild that original B&W 8-bit picture by adding (-30R,+18G,+38B) and then selectively using the shadow detail from the red channel and the highlights from the green or blue channels. The result, in Photoshop, is the same as if the pic had just been shot in B&W to start with (barring JPEG artifacts).

So why does it seem to have more tonality? I'm guessing that it simply looks better on the small, contrasty camera LCD.

It's a tragedy that the B&W image is already clipped to 8-bit before sepia conversion. If the image was unclipped, the sepia conversion realy would be better, with and increased number of shadow tones in red and highlights in blue and green. It is what it is, though.

At least for web use, this is suggesting to me the following work approach for high quality B&W from JPEGs (when time demands preclude shooting in RAW):

• count on spatial resolution as a stand-in for pixel depth detail. Given some level of pixel noise, 4x4 8-bit pixels should be better than one 12-bit pixel. For a reduction from a 4K image down to a 500-pixel web snap, there are a lot of input pixels for every final output pixel.

• with this in mind, set the camera to a low-contrast image, to compact as much of the original RAW range into the 8-bit jpeg, and then feel free to beat on it using levels and curves later on -- knowing that even if precision is lost per-pixel, a lot of it will be made up in the size reduction step.

I'll report on some results soon.

P1010577

P1010333, Alameda

New and Contemporary

There is little that can lead you to treasure good photography than to look at a lot of bad photography, interspesed with an occasional gem. Which is exactly what I was doing a few weeks ago on (where else?) flickr, where I was editing group pools.

When I started the New Black and White group, back in flickr's early pre-yahoo fog, there were no editing or moderation tools, it was slow and painstaking to remove each and every pic I felt didn't belong. And at first that was fine, as there were very few pics submitted. Two or three a day. I stopped messing with it, left it fallow -- came back to find a thousand pictures.

Edited those down to a few dozen, watched it fill up quickly again. Eventually the flood was far more than I could manage as anything less than a full-time job, so I ignored it for months until there were more than 55,000 photos in the pool, most of them "flickr noise" of the cute kitten variety.

Rather than even try to deal with all that, I started another group, Contemporary Black and White, and invited a few select members. I thought: at least I don't need to edit them (and I don't -- they've been contributing good stuff). But then I started wondering about the old one.....

...could I realistically edit-down the 50,000+ photos to a little kernel, throwing away 90% or more of them? Only one way to find out....

So I did. It took time, it took special tools, it took regular attacks. I worked in blocks of from 1000 to 3000 photos. As time passed, it was quick enough to see how many really endlessly-repeating tropes there were, each trotted out over and again and usually accompanied by plenty of enthusiastic flickr badges and boosting comments (especially if the photo involved a pretty girl).

As soon as I started deleting, of course, the hate emails started coming in. What was interesting, if not surprising, was that some of the worst "offenders" -- by that I mean people whose "NB&W" photos were dull, often not even black and white, and massively spam-posted all over flickr -- were the most strident of the protesters. A handful of them diffidently pulled their photos (by the hundreds) from the pool. Thank you for saving me so much time! I thought, but also couldn't help but believe that they genuinely thought they were doling-out a punishment, as if I would somehow be harmed by no longer being able to view (and congratulate) their hundreds of bare-tree-on-the-plains snaps ("Don't bother replying back as I'm blocking you." too).

Photography, especially as practiced socially on the internet, is as vulnerable as any craft to the curious backwards-expectation principle: that is, for any skill, the very best people are constantly self-critical and pushing, always knowing that they could do better -- while the weakest practitioners are quite securely smug in their belief that they've got everything licked, under control, no need to reflect or review except to roll in adulations.

In internet-style photography, this is reinforced to a staggering degree, if only because so much of photography, as generally practiced, is about flattery -- just like most online social networks.

In the aforementioned 1951 Berenice Abbott article, places the blame for photography's emphasis on flattery on proto-pictorialist Henry Peach Robinson, the creator of the famous Victorian melodrama photo "Fading Away," and most especially his book Pictorial Photography (A sample: "It is an old canon of art, that every scene worth painting must have something of the sublime, the beautiful, or the picturesque. By its nature, photography can make no pretensions to represent the first, but beauty can be represented by its means and picturesqueness has never had so perfect an interpreter."). Personally I don't think he can quite be credited for human vanity, but the gun's still a little smoky.

It doesn't really matter where it came from, but the belief is rampant that "good" photography equals "flattering" (this was exactly the criterion given to me a recent local photo class, a class which also extolled the works of Yousef Karsh as expressing the highest of all photographic virtues). It's a flattery that aims both ways: if your pic is flattering, then it is "good" which flatters the photographer.

The cycle starts there: the photos are flattering, and deserving therefore of flattery, and flattery leads to flattery leads to flickr badges and group invitations and should a tiny shred of actual daylight get into the party the guests will be in a ruinous uproar over its harsh direct brightness.

Not that the angry messages bothered me after the first thousand photos. I stopped informing, just went at it deleting. Didn't look at the names, didn't look at anything but the pics. And saw the same two dozen shots over, and over, and over...

I started to catalog them for a while: "...cute cats, adorable soft-toned children, footprints in the sand at the beach, misty time exposures of water, streets and railroads stretching off into the distance, casual snapshots, rusting cars, ruined castles, silhouetted trees, tele shots of resting seabirds, trees and unidentifiable silhouetted figures in fog, photos of backs of heads, (abandoned) factories, tourist destinations, parked bicycles, wide-angle upshots of skyscrapers, cathedral ceilings and windows, photos of camera collections, geometric airport ceilings, exterior stairways, wistful old statues, people looking bored while drinking coffee, interesting architectural details in closeup, sunlit spiderwebs..." it was actually a relief to leave in at least a few flattering model shots, if only because they weren't another empty craggy landscape.

For the sake of completeness (no, just to be random) I kept some in place, throwing away the OTHER photos just to keep runs of clichéd ones: say, thirty black trees on white (URL approximate), or twenty men with hats, or 30 shots of animals -- in order. Moderators can't rearrange the photos, I could only do this by eliminating the pix between how many runs of shots were just the same, same, same.

What none of the protesters did is what would have redeemed them. Despite repeated patient invitations in the face of their ad hominem vitriol, not one of them was willing to say why their photo deserved consideration, why it was special or worth anyone's time to view. Not one.

I can't help but feel the person who has benefited most from the exercise, though the obvious result is a much better pool of photos, is me -- seeing in a deep way, night after night for weeks, just how many "gorgeous!!" photos are really not worth a second glance and how many of them are in the same narrow predictable range that have been little-changed since the 1860's. Now the real challenge: remembering to avoid them myself while still making photos. At least photos I'd bother putting on the web.

Caveat eyeball.

Rethinking Gorman

It’s been a couple of years now since I wrote this entry on digital Black and White conversions. I’m still using a variation of the Caponigro conversion described there. What prompted me here was a combination of events, including reconciling the many scripts and actions I had on several different Photoshop-equipped computers, each of which had diverged from ts brethren; meeting Bob Carnie at Elevator Digital in Toronto, thanks to Dinesh; this APUG thread, which also included more info from Bob; and the latest edition of Digital Photo Pro magazine, which has run B&W articles as its cover story quite a lot over the last year or two, and this one was no exception. What surprised me was that DPP were freshly touting the old Gorman/Holbert method (aka the Gorman Method).

I've tried the Gorman method a few times in the past, stored actions for it like many other people have no doubt, and... it puzzles me. It puzzles me because:

• It uses layers, but it's destructive -- you can't got back to the original unaltered picture, so you'd better be careful not to hit "Save" unless you're really certain that you're 100% done with the pic.
• It's not really a black and white conversion method! It's three different effects, one of which is a black and white conversion -- and a very plain-vanilla conversion, at that. The three effects are:
  1. Black and White Conversion. The Gorman method uses unadorned Lab conversion then propogates the "L" result t all channels before converting back to a full-color RGB iamge. That is, grab the "L" channel and that's your monochrome image. What'spuzzling about this is that "L" is essentially a straight linear conversion, if you work out the math and remember that Photoshop locks the CIE iluminant term to "D50" (so all those spooked cube roots are really just constants). It's a weighted sum of more or less (depending on the color space of the original image). No contrast control, just a straight conversion based on roughly 20% red, 70% green, and 10% blue (again, the exact #'s depend on the input color space).
  2. Color toning. Using the inverted image as a solid-color layer mask is a cute idea for prepping the image for printing or other distribution, and I like the suggestion to keep favorite tones in a palette. An odd part of the DPP article is that Gorman is quoted as talking about using a completely different method of selecting tones for printing -- implying perhaps that this method is only useful for web presentations?
  3. A high-pass overlay layer, based on the resultant B&W image. Not a bad idea, though the exact sizing is somewhat dependent upon the input size, rather than a fixed value as written in the PDF. I also like how te blend has been altered and feathered so that potential "halo" artifacts will be isolated to the midtones, so that they won't appear in the most potentially troublesome parts of the image -- you won't get light halos in black areas or dark halos in highlight areas.
• It makes reference to adding curves or levels layers, but doesn't actually describe where they would be placed or how they would be used. In fact, since some of the scheme is destructive, you can't entirely be sure that the layers adjustments will entirely let you do what you desire, without repeating the later steps (which sadly is a limitation of Photoshop itself, which I'll describe later).

As an Action the method makes sense to some degree -- I can imagine applying it to whole directories of pictures at a time. What it lacks in control it might make up for in volume, though you could say that about any Photoshop/Bridge/Lightroom action.

There's also the proof pudding -- Gorman's published shots. They genuinely look great, but... is it really the conversion, or the practiced studio photographer? If he really uses this method, my guess is that he's learned how to nail his desired B&W results time after time by rigid adherence until it has become very natural to use and light for this scheme.

Bob likes Lab too, according to the APUG thread. I'm still scratching my head, though I suppose that the same "use what works for you" logic could be behind his comments (and I have no argument with that line of reasoning, believe me). It's also true that using a combination of all three colors, regardless of just what that composition might be, has the potential to give you really smooth and gorgeous floating-point-precision gradients even without resorting to Dark Side trickery like HDR.

Still, I like the Caponigro method, and have adjusted it slightly since the previous entry. I don't try to put the color toning into the same operation, and I put all the adjustments into a folder, so I can turn them of and on as a group (or fade them by adjusting the opacity of the entire layer group at once).

I've also added a blank layer I label "burn" to the folder, and set its blend mode to "multiply." Anything I paint into this layer will essentially burn-down (darken) the corresponding areas of the final B&W image, and it too can ave varying opacity in case I get a little too heavy-handed wit the brush (which happens a fair bit). The illustration shows a complete "Modified Caponigro Method" folder.

What about that undisplayed "Layer 2" above the folder?

It's the result of my very favorite Photoshop key commands, a trick I've only heard verbally described as "The Move." It's "ctrl-option-shift-E" and does in one step what the Gorman method does in two -- it merges all visible layers into a new layer at the top of the stack (in the Gorman method, this is done by creating a blank layer and then running "merge visible" on that layer -- exactly the same result).

That "Layer 2" is just such a layer, which has then been hidden for later use (as we'll see).

A problem with Photoshop is that it doesn't let you specify arbitrary blend operations (or filters) on a folder. So the only way to use those sorts of effects are to do a "move," create that new layer, and then execute whatever on that layers pixels. That's what the Gorman method does for its high-pass layer, and how I use it will be described below.

These "merged" layers are great, but do remember -- if you edit any of the underlying Hue, Curves, etc layers, those changes will not automatically appear in the merged layer. Instead, you'll need to hide or delete the old merge and make a new one (clicking the little visibility icons then makes A/B comparisons pretty easy too, btw).

Recently, I've been experimenting with using these merge B&W layers to manipulate color images:

1. Work out the B&W conversion using a folder, as described above.
2. Create a merged layer using "The Move."
3. Hide the entire conversion folder by clicking on its little eye icon.
4. Set the blend mode of the merge layer to "Luminance."
5. Adjust opacity of the merged luminance layer to taste.

Here are some pics: the first is an un-modified original, followed by two possible B&W conversions, and then the color result of applying those conversions back to the original color image. In the first, darker one, I liked the somewhat bad-color-printing appearance it had. In the second, which also washed-out a lot of the detail in the freckles, the color becomes very soft, looking almost like hand coloring. The last in this group uses the same brght luminance source, but the effect is faded back by reducing the luminance layer's opacity.

As long as we're in Photoshop, it's hard to resist trying near-useless tricks, too. Here's a completely negative luminance applied to the original colors, and its inverse (which gives positive luminance to negative colors).

These samples use a different blend mode: the "lighter color" mode. The result (which looks best, IMO, when slightly faded back) gives slight variations in the overall color saturation according to the overall luminance -- another rather film-like effect and one I really like.

Finally, why not blend components of techniques for something new? The pics below show the high-pass layer from the Gorman technique, applied via "overlay" to the original color image.

So what about video games?I would be remiss not to mention that all of these methods, except for the high-pass filter, can be executed in a single unified pass in hardware shaders, thus also making them appropriate for using in video games (the high-pass filter would require an additional render-to-texture pass). If you've been paying attention to John Nack's blog, you may also know that such effects will eventually be available in real time for Photoshop and Flash via "Pixel Bender" shaders.