Pattern Recognition

 Mara Sedlins:

This week I’ve been busy preparing finders’ prizes for all of the Cerebus Art Dragnet contributors we have so far. The current tally is 113 original art scans from 32 contributors (excluding a few who declined a prize). Thanks again for helping us gather these scattered pages!

The other main dish on my plate is the C & S I Trauma Pages, those original art scans flagged for special attention. So far this has meant focusing on the more complex types of tone that have shrunk or torn.

Above are a couple examples of what I mean by complex tone (I think of the first one as the “wiggly tone”). Do you recognize which character the second tone was used for?

Replacing Cerebus’ tone has become pretty routine - it usually works to copy and paste a patch of repeating dots from a nearby area, line it up with the neighboring dots, and then erase anything that extends past the repaired section. But with the textures above, there isn’t a simple repeating pattern that we can identify and leverage in the same way. Instead, we use one of several techniques depending on the size and shape of the missing area. My favorite is “drawing” in the texture with Photoshop’s stamp tool, being careful to blend the new shapes with the existing tone in a consistent way. A crucial last step is printing out the restored page, squinting, and looking for telltale lines that reveal where the fix was. If I did a good job, the tone is seamless. If not, it looks like this:

(Note: this is an intentionally poor job, for illustrative purposes!)

It’s harder to achieve a perfect fix with the second type of tone shown above, because the pattern is less random. If we could identify the boundaries of the repeated spiraling pattern (assuming there is one), we could use the same technique as with the Cerebus tone. But neither Sean nor I have been able to do this by visual inspection.

I did some research into computer algorithms for detecting repeating patterns in an image, to see if there’s a way to automate the process. It seems like a simple problem - but based on the number of academic papers I found and the variety of different approaches used, it looks like it's still a nontrivial issue in computer vision research. In terms of actually implementing a pattern recognition technique, I’m not sure where I’d even begin (I suspect I’d need to learn Python and read this book, and/or delve into open source code - probably not the best use of my time for actually getting the restoration work done, especially since the tone in question only appears a handful of times.) If this is anyone’s area of expertise and you have a solution, though, let us know!

Since my last update I started reading C & S I, from the beginning, for fun (imagine!). As with High Society, even after cleaning over a hundred pages, my sense of the plot is going to be pretty fractured before I sit down and read the book as a fan - restoration is just a totally different way of perceiving the pages.

I did get a sense of the overall structure as I flipped through each page doing a quick catalog of “icicle” text bubbles, following up on Jeff’s comment on my last post. Can you guess who speaks with that tone the most often?

(drumroll .........)

Okay, maybe it’s not a surprise since he’s the titular character, but Cerebus comes in first with 28 chilly statements. His icy tone is directed equally often at Jaka and Weisshaupt (six times each). The second coldest character is Bishop Powers, with 11 wintery phrases, spoken mostly to Weisshaupt. I’m beginning to see which conflicts are going to drive the plot ...

Meanwhile I'd love to see more finders' prizes go out soon, so keep the original art coming in!