To generate “clean” averages, then, I would need to sort the QRs by size and produce averages off those subsets. Which is what I did. Here’s a quick histogram of the QR sizes I have, using today’s set of 180 total QRs:

This is something of a “long tail” graph, with a big bulge in the smaller sizes and very few in the larger sizes. In fact, the three largest sizes only have one member each, which means no composite is possible. Those I will not include. The rest, however, I did composite, and the results are here:

216 pixels:

248 pixels:

280 pixels:

312 pixels:

344 pixels:

376 pixels:

408 pixels:

Notice how crisp the anchoring areas of the QR graphic are now, as opposed to the data areas, which vary with each QR. Alas, none of these “clean” composites actually goes anywhere, at least not with the QR reader on my phone.

And here are the as-shot and generated QRs, numbers 151 through 180:

And the composites from the entire set of 180:

After doing this, I thought of a post I’d done long, long ago about extending the English alphabet and coming up with some new letterforms. Nothing much came of it, but I’d thought of doing a New Alphabet sampler then, so I thought, why not try something like that now? Instead of coming up with new letters, though, just rearrange the letters in a new way and try to shoehorn the extraneous ones in somehow. So, here’s the Articulation Alphabet Sampler:

This particular sampler arranges the letters in order of place of articulation, ranging from P and B using the lips all the way back to H and X using the back of the throat–in this case, “X” standing for the Germanic sound you get in “Bach”. The other oddball letters are C, standing in for the “ch” in “church”, and Q, which is supposed to be sounded with a forward “rk” sound in words like “Iraq”.

You can even sing it:

♩Pee bee em ef vee eye tee ♫
♪ Dee ee ess zee en ell sea ♩
♫ Jay are ey ♩
♩Why cue kay ♬
♫ Jee you double-you oh aitch ex. ♫
♩Now I know my P-B-Ms ♬
♪ Tell me what you think of them. ♫

For this end, though, I’ll need to modify the header definitions. Consider this ETG V0.2, with a small tweak on how to deal with images that are supposed to be displayed together.

The ATG frames code includes a 4-letter ID squib and 2 hex digits for animation frame numbering. I reserved the first block of 128 for a non-looping animation set, which is actually kinda silly because most Web animations are looped anyway. So let’s use that block to indicate the location of each tile in the set. This gives us a latitude of 7 tiles across and 15 tiles down, or tile coordinates 7E.

If you have a tileset of 7 tiles by 10 tiles, then the tiling number 45 means “fourth tile from the left, fifth tile down”. The tiling number 12 means “first tile from the left, second down”. And, using hex, number 7A means “seventh tile from left, ten down”, or the last tile in this set.

For this image, I tiled an image into a 3×3 grid, or 9 tiles. For our first tile, the header will look like this:

`CGA which gives the color space

`09×12 which gives the size of the tile

`FedW11 which gives the ID squib and the origin tile (first from left, first down)

and `{ which locates the data payload.

The tiles can be tweeted in any order, because they share their unique ID squib and their ordering is contained within their headers. However, I prepared these in left-right/top-down order, so I’ll paste them here that way:

`CGA`09x12`FedW11`{000600000000000000000400000000400000000004000
0E7E80000000000000000000000000000000000000000000000000000000000

`CGA`09x12`FedW12`{000000008000000067000688087000C7707700077C867
000C7786700087778700067787800008E878066608887086206878600088687

`CGA`09x12`FedW13`{600068808680087808600066888088187780938077778
997067787199077777039187C760091177770001199E7000009993000000399

`CGA`09x12`FedW21`{006008604060666660604006086000040660000000004
0000000040000000000000000000000000008800000088860000870E668807E

`CGA`09x12`FedW22`{E777FFF776776877EF788888667766886877767888677
7777767F777E77777E777E77F776777F77777C67777688777C7788787C7C7F7

`CGA`09x12`FedW23`{677E6CC8867C7777888677E7777066777F67006767777
800087877780000660668804008778888860777788A007778A6630137787710

`CGA`09x12`FedW31`{0608000686000600460400060064008EFFF70000666E6
40006EE6A0000680600646066EE0600460666664008EE7C760646E7766800EE

`CGA`09x12`FedW32`{EE77A997E7F77708E6F7F7E66ECF76866F6F7C666A6A6
77088666A7E76E0000F77E7E8677FEEE6000777770444066E6000666E660000

`CGA`09x12`FedW33`{C76E00000C676400006E76040047670404067E6040040
780004000600000000000000000000000000000000400000000000000000006

Here are those pieces:

Here they are assembled:

And here they are enlarged for more clarity:

One of the easiest things to see in the data are strings of repeating elements. If I could further encode those strings I could save room and not have long lines of 000000 or 33333 for fields of color in the image. Let’s say I simply put the number of pixels in front of the code for the pixel color, and the TwGD will decipher what was meant. So, a string of eight white CGA pixels, FFFFFFFF, would be represented as 8F.

The only problem with this is that the character 8 already encodes for dark grey. Putting numbers in the stream will collide with the numbers already there. So I’ll need to convert the hex numbers into a new coding scheme, so the pixel quantities don’t mess with the pixel codes. Say, pixel 0 (black) equals a, pixel 1 (dark blue) equals b, and so on. In this scheme, the string of white, FFFFFFFF, would become 8p. That’s a savings of 6 characters!

Of course, you won’t get that kind of savings everywhere. Especially 2-pixel doubles, since you’re just replacing 2 identical pixel codes with a “2″ and the new code. There’s still some squishing to be had here. For this special case, I came up with a second series of codes, using punctuation to stand for pixel doubles. Here’s the full set of codes for this scheme, which I’m calling Similar String CGA encoding (“CGS”)

0 = a   00 = !
1 = b   11 = "
2 = c   22 = $
3 = d   33 = %
4 = e   44 = &
5 = f   55 = '
6 = g   66 = (
7 = h   77 = )
8 = i   88 = *
9 = j   99 = +
A = k   AA = ,
B = l   BB = -
C = m   CC = .
D = n   DD = /
E = o   EE = :
F = p   FF = ;

So, using this additional compression scheme, I was able to encode a 12 x 17 pixel CGA image with plenty of room to spare. It doesn’t sound like too much extra, but that’s 150% the area of the 10 x 13 CGA pixel image I tweeted back in December, and includes the ETG header information, too.

Here’s the image, a color version of the self-portrait from yesterday:

And the resulting tweet, using “CGS” as the code for this new compression scheme:

`GCS`12x17`{6ag12ae!ghi9ago11ag8aigag4ah*gpiag3a4h3pigahih3ihphioa3hp
)phigia)ih;hp)!*)p3hpi3ai5hgh3a3ihp4h!*haihp)g!b)i!i)3ajd)i8abj3hi5a

`2bg`15x25`{K5LLX01MH5G41Kle00GR4004P0006M01GQL07glb0R$$r9UQ$lUkckhlh
$!PFl$$zMh$luck$gq6R!hm6Q!gWRMl$Hdql!LNj1O0Nwb0G2lgG51hxH006fGH00004

And the encoded image:

HIE will record visible light, too, so I used a 780nm filter, which looks just like a piece of black plastic. I taped the filter to the lens so I could flip it up out of the way to compose the shot, then flip it back down to take the picture. I never knew exactly what happened when the shutter clicked until I got the film developed.

Here are the two images that were accepted into the show: “Globe & Window” and “Vent & Yucca”. I may post more images later on.

Batch 05 as-shot:

Batch 05 generated from the photos and composited:

Full set of 150 images as-shot:

Full set of 150 re-generated: