The Week in Pro Photo Software

For lack of a better title this post gets what it gets. There have been a number of things going on in the past week in the area of RAW and pro photo issues. This week was Microsoft's Pro Photo Summit. Wish I could have been there. Through various related link hops, I stumbled onto Bill Crow's Media Photo Blog. He's Program Manager for Windows Media Photo, Microsoft's new still image format. Also, this week a couple of acquisitions: Microsoft purchased iView and Adobe purchased Pixmantec (as you can see from their updated web pages).

Colin Hay

I've long been a fan of Men at Work front man Colin Hay, both with the band and solo.



Today, I stumbled onto this excellent classic bit from Scrubs with Zach Braff and Hay singing Overkill.

The Standard: Ed Catmull

Stumbled upon this article about Ed Catmull in The Standard, "China's Business Newspaper." I found it interesting to see this covered in an English language business paper based in China. I enjoyed the article, and I really like the Ed-isms, with which I agree strongly.

"Pixar people frequently recite Ed- isms:

Ed believes that you should always hire people who are smarter than you.
Ed believes it's more important to invest in good people than good ideas.
Ed believes that you learn by making mistakes and that success often disguises problems.
Ed believes that magic happens when you don't operate out of fear."

link

Before 2046, Aimee Mann & Wordplay

Three things have engaged me in terms of entertainment over the past week.

We watched Kar Wai Wong's In the Mood for Love (Fa yeung nin wa) the other night, largely because it's a prequel to 2046, a film on my list. I can't give In the Mood a thumbs up or down at this point; in spite of interesting cinematic elements, I fear I wasn't in the right mood for its subtleties.

We saw Aimee Mann play the other night at the zoo. The playlist included Save Me, Lost in Space, That's Just What You Are, etc. Having seen her a couple of times, I've come to the conclusion playing Voices Carry is out of the question, but I still really wanted to hear it.

And last night, we saw Wordplay, "an in-depth look at The New York Times' long-time crossword puzzle editor Will Shortz and his loyal fanbase." My wife and I both enjoy doing the Times' puzzle (although at this point I'm feeling terribly rusty). We love ethnographies, the film was much enjoyed; it was well done and included interviews with crossword fans Bill Clinton, Jon Stewart, Ken Burns and others.

If you really want to geek out on word puzzle movies, you can make a double feature with this one and the 2004 Scrabble player ethnography Word Wars. For a triple feature, throw in the excellent 2002 documentary on National Spelling Bee contestants, Spellbound.

Ze is Amazing

Boing Boing recently tipped me off to Ze Frank's hilarious vlog, The Show. I've linked to Ze's site a couple of times before, and he's really hit a home run this time. He is an amazingly talented guy.

This is another good example of what happens with the democratization of media. When it's possible to bypass corporate bureaucracies with layers of middlemen and broad distribution is easy and relatively inexpensive, new stars continue to pop up seemingly out of nowhere.

The disruption seems to be still just beginning.

link

Video Memory Lane

Hardly breaking news, but I realized YouTube is currently host to practically every music video ever made. The other night, this realization translated into video reminiscing into the wee hours. Of course, A-ha's Take on Me still is the greatest music video ever made. Chances are these will all be sued off the Net in time, so enjoy them while you can.

Freakonomics & Hidden Order

Freakonomics came in the form of a Father's Day present, and I promptly reclined in a lawn chair and absorbed it in Sunday afternoon shade. An enjoyable read, "rogue economist" Steven Levitt explores "the hidden side of everything." Topics include the financial structure of inner city gangs, cheating at school and sumo wrestling, statistical significance found in baby names, etc.

Although Freakonomics generated much more buzz, it reminds me very much of a book published ten years ago, David Friedman's Hidden Order.



Friedman's book is lengthier, and it gets into more technical detail, marginal value, supply and demand curves, explanations of comparative advantage, etc., but it contains many of the sorts of economics analyses of everyday life found in Freakonomics. If you liked Freakonomics, Friedman's book is worth a look.

GPGPU, MIT 6.098 & 6.882, OpenRaw

i. From the GPGPU workshop at ICCS 2006, here's a link to a tutorial on Scientific Computing on Graphics Hardware.

ii. Computational Photography Update: M.IT.'s Fredo Durand's page is worth a fresh visit. His syllabus pages for 6.098 Digital and Computational Photography and 6.882 Advanced Computational Photography are excellent.

iii. A while back rawformat posted a link to a recent OpenRaw survey offered to the professional photo community regarding digital camera RAW related questions.

Even Educated Fleas...

Yet another shot I took in Italy, if I haven't already posted it here.

The Eyes

My brother and I have been going back and forth on the random tile question (1, 2, 3). Related, at least in my mind, is an information theoretic question dealing with what see see and how much the optic nerve is capable of transmitting (as best we know of our own mechanics and the principles involved). Perhaps I should research this more (and find the source), but take, for example, this comment from Robert Cringley in 2004:

"What we need to emulate here is the eye, itself. Look at the optic nerve that connects the retina of your eye to the visual cortex of your brain. The optic nerve is composed of approximately one million stringy cells called ganglia that fire in parallel over a distance of two to three centimeters with the actual visual signal travelling at about 4,400 feet-per-second. Taking into account recovery time between signals, the optic nerve has a total bandwidth of approximately 100 kbps..."

Let's suppose that 100kbps for the optic nerve is in the ball park (even within an order of magnitude). The next obvious question is the information content of a single frame of what we see given our visual accuity. Compare what you see with, for example, a 10 megapixel image.

Comparisons I've seen are on the order of a 100 megapixels per frame (HDR, of course). Given the amount of information we see in a single frame, the number of theoretical frames per second we see and the supposed diminutive nature of the pipe through which it's transmitted, something doesn't add up.

Is the brain doing a whole lot of synthesizing and hallucinating with the little data it gets via the optic nerve, is there some incredible feature extraction, encoding and amazing compression going on somewhere in the process, both? What's going on?

The thing about noise is that it has high entropy as its defined in information theory. Gaussian noise has the maximum possible entropy (e.g., information on distribution entropy here); this is a factor contributing to my surprise at the results. It leaves me asking questions about how our human visual systems respond to data that presents a serious encoding challenge in light of Shannon, what gets passed through to our brains, etc.

Googlements

Google launches Picasa Web Albums, which is guaranteed to invite comparisons with Flickr.

In Google Fatigue Sets In, Wade Roush at Technology Review writes: "Users are reacting to Google's new online spreadsheet with a big yawn. Is it a company searching for a strategy?"

But I think Nicholas Carr has it right in considering Google's spreadsheet an Office add-on, saying, "So why would Google put out a product that makes its arch-rival's product more valuable? Because Google doesn't want to compete with Office. It sees Office as part of the existing landscape, and it wants to build a new layer of functionality on top of that landscape."

Also via Carr, the NYT piece "...Google seeks more power" details Google's building of a massive computing center in Oregon on the banks of the Columbia River. "...it is hard to keep a secret when it is a computing center as big as two football fields, with twin cooling plants protruding four stories into the sky."

In other news, I think I read something about Scoble leaving Microsoft. Fafblog's satirist extraordinaire Giblets strikes gold again. And, an old educational film: Powers of Ten.

Don't You Know Who I Think I Was?

In 1997 when The Replacements came out with their greatest hits, All for Nothing / Nothing for All, their earlier work from the Twin/Tone label was conspicuously lacking. This time with Don't You Know Who I Think I Was - The Best of The Replacements, that's not the case; for example, you'll find I Will Dare, a song from Let it Be that always deserved inclusion.

(I am going into insanely busy summer mode, so posting may be sporadic)

Colored Castle

The name for this blog came rather spontaneously. I was reading about color science--metamerism specifically--and since it made no sense at all to turn the -ism in to an -ist, I figured there'd be a decent chance of certain degree of Net-uniqueness to it.

That was supposed to segue into the following illusion, but I couldn't make it work, so I guess that paragraph is nothing more than free minutia. Via 0xDE, comes a link to this cool perceptual illusion. You stare at a dot in the center of a picture of a castle, move your mouse over the image, it changes to grey but appears to be in color for quite a while if you don't move your eyes.



It's quite good.

Check it out: here.

Note: I took a stab at rolling my own and had pretty good results, although I'm sure they can be improved.

My first try:

1. Split image into H, S, L channels.
2. Duplicate L channel.
3. Set L constant (I set it at 128/255, but that was probably too much).
4. Recombine H, S, L.
5. Invert it.
6. Paste image from step 5 over duplicate L from step 2.
7. Put little black dot in center of inverted color layer.
8. Make top layer invisible revealing duplicate L.

Do You See What I See?

Given the last tiling in my previous post with an image produced from [allegedly] true random numbers from random.org, I circled the most prominent features I perceive in the image, what to me is a repeated lighter area. I did this quickly and sloppily, but the regularity of the resulting lattice is still quite apparent.



My current hypothesis is that there is some perceivable order in the "truly" random pixels; that is, there's some sort of statistically significant difference between the patches I selected and the rest of the tiles.

If there were no order I think the best I could do is receive impressions of repetitions of the whole tile in its entirety. There's a possibility that it's an optical illusion--i.e., the brain can't do anything with the randomness, so it consistently makes something up. This is why a good statistical analysis is in order (if time and curiosity go my way).

I also believe there's a high probability of this happening with any random tile (a high probability of some perceivable order in a tile this size). A good path for further research is trying to find a random tile that produces no perceivable patterns.

Mark was right in guessing that this isn't a problem with the random number generator as much as the brain's ability to find order. How we do it, and how we do it so quickly is, again, really amazing. Consider for a second the amount of processing this might require if done programmatically.

A related question asks if our visual processing systems do some sort of Fourier analysis or something equivalent. I find the latter option the most interesting. It's hard to imagine our brains doing true Fourier transforms, so what's the equivalent? (Then I worry who will attempt to patent it first--argh!).

Google += Andrew Moore

I'm by no means a machine learning expert--just a humble soul working at getting up to speed in the area--but I'd still like to note another ML expert being scooped up by GYM. This time, CMU's Andrew Moore goes to Google. According to his Statistics Data Mining Tutorials site:

"Advertisment: I have recently joined Google, and am starting up the new Google Pittsburgh office on CMU's campus. If you're an expert programmer or researcher with an interest in machine learning, please send me email: awm@google.com!"

Also, I've noted it before, but if you're interested in machine learning and related subjects his tutorials and slides are very good.

More Noise

This time using random numbers from random.org.





I still see periodic patterns, but, again, maybe this should be double-blind.

The human brain never ceases to amaze me.

"A huge fraction of our brain is devoted to vision. One of the neglected features of our visual system is that the raw image falling on the retina is severely blurred: while most people can see with a resolution of about 1 arcminute (one sixtieth of a degree) under any daylight conditions, bright or dim, the image on our retina is blurred through a point spread function of width as large as 5 arcminutes (Wald and Grin, 1947; Howarth and Bradley, 1986). It is amazing that we are able to resolve pixels that are twenty-five times smaller in area than the blob produced on our retina by any point source."

David MacKay, Information Theory, Inference and Learning Algorithms, p. 565

The Nightfly

Steely Dan's Donald Fagen was nominated for a grammy in 1982 for his The Nightfly. Toto won, but I think Fagen got robbed; over the years this disc has retained its position on my favorites list, and I still consider it one of the best discs made in the last 30 years.

It's a tribute to growing up in suburbia in the Eisenhower years. Before my time, but hardly beyond my appreciation. There's the marimba. I.G.Y--International Geophysical Year, the bittersweet, sardonic voice and ironic lyrics dedicated to the overly optimistic 1959 visions of a Jetsons future side-by-side with Cold War visions of dancing with a beautiful blonde in a backyard bomb shelter.

"She loves to limbo, that much is clear. She's got the right dynamic for the New Frontier."

The Probability of Order

In the previous post, I talked about the problems I ran into tiling random noise and how it produces patterns. The first possible caveat might be a flaw in my random number generator. Maybe I should go back and get data from random.org, but I don't believe that will make the problem go away. Another problem with all this is that it's not double-blind--I really should have someone else blindly deduce the tile dimensions of the resultant matrix.

One thing interesting regarding my own perceptions is that I see patterns in the noise, and that's the problem. I see a repeating dark spot of sorts and sort of a lighter swoosh beside repeated four times in both dimensions. If I had time, I'd circle these to spots and compare them with the general population of pixels in the tile. I'm willing to bet there will be a statistically significant difference.

Please excuse all the speculation, but raising questions regarding randomness, order, perception and how the brain works is a much greater objective here than finding answers.

My conclusion, and it very well may be wrong, is this. Given a tile of random pixels, there's a high probability of it containing some sort of perceptible order--e. g., perceptible enough to make tiling random noise an unworkable means of generating a significantly large image that appears random.

A more general (and perhaps too general) question is this:

Given randomness, what is the probability of order?

So far I've yet to run into this question or any related answer, but it may simply be the result of my own ignorance. If you stumble onto this post, and you have any good pointer, comments are most welcome.

Order seems really hard to quantify. One can calculate the entropy of a distribution. Intuitively, I suppose one could set some threshold for the entropy of outputs and calculate the probability given the distribution of the inputs.

On the way home I started thinking of a 3 x 3 grid in which each cell could be colored black or white. There are 512 ways of filling in such a matrix (2^9). If we define order as a three unit straight line, then we have 10 possibilities (3 horizontal, 3 vertical, 2 diagonal). Given all that, the probability of order given randomness is 10 / 512. That's a simple random system with some definition of order and a probability of order occurring, as it's defined.

I'm left curious and pondering how the brain processes visual information, how it finds patterns and order (even out of mathematical randomness).

C'mon See the Noise

A few years ago, I was working on writing code to rapidly create an image full of random noise. On the first stab, I created a small tile of random noise like this:



Next, I tiled it to create a larger image. For example, here's the tile repeated four times in both dimensions.



To my chagrin, the results always contained discernable patterns. If you look carefully, you should be able to see what I mean in the example above.

Intution told me, wrongly, that it shouldn't happen--after all, what's being tiled is RANDOM (well, pseudo-random). Yet, still, some how consciousness is able to find the patterns.

Why does this happen? I have come up with one explanation, and I'll try to go into it tomorrow.

The more I know of information theory, digital signal processing and consciousness, the more fascinating this is to me. We're not done learning in these areas.

Finally, the fantasy billion dollar idea. ;)

A large proportion of our brains is devoted to the processing of visual information. Is there a way to encode, say, financial market information into a visual form in such a way that we can harness our advanced visual processing capabilities to find important connections and correlations?

This might be a job for the Mechanical Turk. :)

Koi

The Price is Right

When it comes to setting a maximum bid on eBay, I tend to choose numbers such as $20.01, working from the assumption that most people will choose rounder numbers; i.e., $20.01 beats $20.00. Of course, maybe $20.02 is better because it will beat people with a $20.01 strategy. The optimal strategy? Therein lies the problem.

Major Time Out

In case you missed it, the big news in biology this week was the Israeli discovery of an underground ecosystem millions of years old.



"Discovery of eight previously unknown, ancient animal species within 'a new and unique underground ecosystem' in Israel was revealed today by Hebrew University of Jerusalem researchers...

The cave, which has been dubbed the Ayalon Cave, is 'unique in the world,' said Prof. Amos Frumkin of the Hebrew University Department of Geography. This is due mainly to its isolation from the outside world, since the cave's surface is situated under a layer of chalk that is impenetrable to water. The cave, with its branches, extends over some 2.5 kilometers, making it Israel's second largest limestone cave. It is to remain closed to the public to permit further scientific research."

more at the Hebrew University page.

Before the Idol

As far as it goes, I don't watch much television. Some popular shows I've seen twice or once or never. Such is the case with American Idol.

From listening to others who watch Idol, I've gathered one of the appeals of the show is Simon's brutal smackdowns of the talentless. This is an aspect of the show I've found less appealing the few times I've seen it, but the structure of it leaves me thinking of The Gong Show of my youth.

Before American Idol, there was Star Search, and before Star Search there was Chuck Barris' Gong Show which took itself far less seriously. Even when it was brutal, it was tongue-in-cheek, and probably a set up.



The show consisted of Chuck Barris, three stars (usually Jaye P. Morgan, Arte Johnson, Jamie Farr, etc.), a large gong, and talent show contestants who attempted to perform their acts without getting gonged.

Through the magic of iFilm, Google Video and YouTube, I've found a few old clips of the show. On YouTube, Gong Show Juggler is an example of an enteraining act going ungonged. On iFilm, there's an entire Gong Show page with about a dozen clips, with Mick Donnelly's Raindrops Keep Falling on My Head being the best example of Gong Show brutality.

Transitional Bopability

Hoping to catch up with a few thoughts here and there this weekend, but there are plenty of unpacked boxes waiting for me.

Anyhow, the first thought has to deal with the art of making a good mix tape (of course, now it's a "disc" or "playlist" but old habits...).

Half the battle in making a good mix "tape" involves making the right transition from one song to the next. It needs to flow naturally as done by a DJ.

So far, and maybe I haven't been looking hard enough, I haven't heard of any of the online music services attempting to collect this information and utilize it.

The question of the day leaves me pondering for a way to successfully mine this data from user listening habits. Done right, it might improve automatically generated personal radio stations offered by various online services.

1pointsize

Recent meanderings took me to a few Indian ad agencies. I love the following ad for Kamdar Opticians by Sharad Haksar's Chennai-based 1pointsize. (Here's a link to a 2004 article on Haksar in The Hindu Business Line.)



On a related note, check out Arvind R's ...How Advertising Spoiled Me..., an interesting Chennai-based advertising blog.