Sean Michael Robinson:
Another quickie this week. I'm in process working on the layout and cover of Going Home, and, since I'd yet to write about or do any de-screening work on this project, I thought it might be an interesting topic for some of you out there.
As Dave mentioned in a previous update, the original photographic slide that was the basis for the cover of the Going Home trade as well as the first issue—taken by Gerhard "from my driveway at "Camp Woolner"—has gone missing, most likely, for good. What this means in practical terms is that the cover has to be recreated from the materials available. With the Reads cover, I was able to significantly improve the image in the reconstruction, because all of the original materials were available and so could be digitally scanned and placed without losing any generations, using the original cover only as a guide. In this case, I can recreate the logo and text, but the image itself has to come from a print copy of some kind.
Why, exactly, is this a problem? Haven't I been working primarily with print copies for the Cerebus Volume One restoration, with excellent results?
The Going Home cover, like you've Never Seen It BEFORE—WAY TOO CLOSE!!!1!
The problem is the printing method. Almost all commercial color printing is produced via some kind of half-toning, i.e. printing with an array of tiny dots that vary minutely to create the illusion of tone across the image. (The effectiveness of this illusion, by the way, is dependent on the printing surface and pitch of the dot, viewing distance, and the visual acuity of the viewer.)
So, if you're printing from a second-generation color image, i.e. an image that has previously been half-toned, you have to get rid of any sign of those previous printing dots, or run the risk of generating the dreaded moire when your previous dots get resampled into brand new dots when you print. Worse yet, you want to eliminate the previous dots without eliminating all of the detail that's apparent in the original image. And it's this part that's the real tall order.
The image above is a close-up of the cover of my copy of Cerebus 232, which also features the Going Home cover image, scanned at 1200 ppi on my Epson 10000XL. I've scanned it way higher than you'd normally need a color image because a few descreening methods (and there are several) work best when the printing dots are resolved, i.e. are reasonably round in the scan.
In this case, though, I found that the fancy-pants methods were not necessary. Photoshop's Median Noise filter did the trick.
In Photoshop, go to Filter-> Noise -> Median. The Median filter looks at the luminance of adjacent pixels to create a kind of average at a user-determined pixel radius. In this case, moving the radius to 8 px resulted in this image—
You can see that most of the "noise" of the printing dots is eliminated, leaving a color cast that's much closer to the original image. And although the detail is still soft, I can see on-screen every detail that's visible to my eye in the original image, including the heavy grain from the enlarged slide.
We can take care of the remaining noise in two more steps—first, downsample the image to the resolution we're actually going to be using it at.
Here's the image downsampled to 450 ppi, using "Bicubic smoother" as our transformation method.
And there's still just a hint of the direction of the original halftoning, so I'm going to finish this off with just a bit of Gaussian blur, at 0.6 px across.
Okay, now that we've successfully eliminated our original printing dots, we're free to treat this image just like we would any other raw color scan, albeit a rather soft one of a fairly noisy surface. We can sharpen, adjust, color-correct, whatever adjustments will bring us closer to the printed original.
In my case I ended up abandoning my print scan and working from a scan Sandeep sent me instead. Dave found the match print of the original cover, superior to all of the other sources available as it includes the bleed area (the area overprinted and trimmed off to create a full-bleed image) and doesn't have folds, spine text etc that would otherwise need to be corrected and touched-up.
Above is the raw scan of the match print. The exposure is a bit dark and the colors seem slightly off to me, but it's otherwise looking good. I notice there's some damage under the logo, caused by (possibly) buckling of the laminate? It looks like little streaks of light.
That, combined with the fine text that I don't want to have to take into consideration when I'm adjusting the photo, persuades me I might be better off eliminating the logo and text entirely and adding them on again after the cover restoration is complete. We'll see.
The Median filter and Gaussian blur combo works perfectly. Then it's on to the cleanup.
I do some of the cleanup to the damaged area of the print, using the Clonestamp tool to sample nearby unaffected areas and varying the brush to hide the work. This is much harder than it normally is, as the visible photo grain makes any blurry edge work visible. After a bit I decide that, yes, the logo should also be eliminated and added back in when the work is complete—the sharpness has been too diminished by the noise reduction. But before I do any more cleanup, I make some levels adjustments and color corrections (as corrections layers that can be changed at any time) to get a little closer to the final appearance of the image.
And here is the color and levels adjusted image, with a rough (rough! so spare me the comments :) ) approximation of the logo elimination. Next step will be to finalize the work on the logo, drop it into a layout program to add back in text, bar codes etc, and print a few samples to take a look at the adjustment work. Then I'll send a few versions out to Dave for his approval, of the colors particularly.
Any thoughts or questions? Hit me in the comments!