Check out the Sudoku Magic iPhone app from Magic Solver. It uses computer vision to detect and then resolve a sudoku puzzle embedded in any picture.
Here’s an example, clipped from my Saturday newspaper:
I used my iPhone 3GS to take the above photo. Next, I launched Sudoku Magic and told it to capture a sudoku from this photo. For a human, picking out the sudoku from this image is a piece of cake. But think about the problems you’d need to solve if you were going to write code to fish out the numbers and number placement automatically.
You’d have to suss out the border of the puzzle, identifying the bounds and correcting the skew of the puzzle (I held the camera off at an angle when I took the picture). You’d have to detect the 3×3 boxes as well as the individual boxes within each 3×3. You’d have to filter out the noise in the photo and detect the numbers in squares with numbers. Each of these tasks is daunting and requires some pretty sophisticated mathematical analysis.
Sudoku Magic does an amazing job. Even with this skewed, low contrast image, it was able to properly identify the puzzle and correctly identify each of the numbers. Here’s the results of this particular effort:
As you can see, the puzzle pulled from the original photo matches perfectly. Now it’s time to play. And, if you get stuck, you can press the Solve button, but that’s kind of cheating, no?
Sudoku Magic is terrific, deserves 5 stars, no doubt. But I did find a bug in the version I tested. I took a screenshot of a web-based Sudoku puzzle and took a picture of the puzzle on my display. This resulted in slight Moiré patterns on the image I submitted to Sudoku Magic. No matter how many times I tried to retake the picture, those pesky Moiré patterns proved too much for Sudoku Magic.
When I tweeted my problem, the folks at Magic Solver were quick to respond and promised a fix very soon. Score another 5 stars for customer support!
I love this app. It does so much more than simply jumble together various iPhone interface elements. There’s complex science at work here. Bravo!
This is a member of the Carnegie-Mellon University robot soccer team:
To get a sense of scale, that orange ball is a golf ball. The robots are autonomous, meaning there’s no remote control involved. They play as a team of 5, with individual thinking, as well as the ability to think as a team. They use adaptive intelligence (they learn as they go), are location aware, have fast vision and excellent mechanics.
To get a sense of this incredible technology, take a look at the video below. Best bet is to view it full screen. The ball moves incredibly quickly, but each shot is followed by a slow motion replay.
Thanks very much to my buddy Todd for sending this my way. Daniel and I are heading to Carnegie-Mellon in the morning for an interview. Will try to get to see these guys in person, take a few pics.
Here’s the CMU robot soccer web site. This particular page is for the small-size soccer robots. Explore the rest of the site to see the dog robots. Very cool!
Stephen Wolfram is a lot of things. The common threads would have to be mathematics and genius. Here’s his wikipedia entry. His main claim to fame is Mathematica, a piece of software that can do just about any math computation you can throw at it.
Wolfram’s latest project is Wolfram Alpha, a search engine that combines natural language processing with expert guidance over a limited search domain. In plain english, ask Wolfram Alpha your question and, if it happens to have a good data source on that particular topic, it will present a customizable framework that answers that question.
Wolfram Alpha excels at any question that allows or requires computation. You can type 2+2, that sort of thing, but that’s too easy. Here’s a bit of introductory calculus:
There’s not a better mathematical resource available, and Wolfram Alpha is free.
What else can it do? Lots! For starters, take a look at this short video. Or, go to the site and type in your own queries. For example, try “gdp of france” or “melting point of copper” (quotes not necessary). There is a bit of a learning curve you’ll have to bust your way through but, if Wolfram Alpha knows about your search domain, it becomes an incredible resource.
From Stu: IBM put together a marketing page to promote Watson, the artificially intelligent system specifically designed to compete against humans playing Jeopardy. The focus on the Watson project is not to win big cash prizes, but to demonstrate real progress is the quest to have a computer system answer questions posed by humans using natural language. Click here to read about the DeepQA Jeopardy Challenge, and click here to read a bit more about the DeepQA Project.
Found this article on the New York Times site. In a nutshell, IBM is pushing the boundaries of artificial intelligence by creating a piece of software that will be pitted against human Jeopardy contestants. A bit of a holy grail for AI enthusiasts.
Be sure to keep the playing field level, IBM. No net access for the software, right?