Slashdot Mirror
Cell Phone with Camera = Scanner
An anonymous reader writes "TechJapan has posted a translation of an Impress Watch Article regarding a new technology developed by NEC and the Nara Institute of Science and Technology, that lets people use their cellular phones with cameras as scanners. It says all you have to do is move your phone over the surface of the piece of paper while recording a movie, and the technology (some sort of software I presume) will construct a high resolution image from the individual frames of the video.
Here is the original (Japanese) NEC press release." I'd love to see before and afters to see how well this works.
Whether some trick like this makes it happen sooner rather than later only time will tell, but eventually just in terms of raw resolution camera equipped cell phones will be functional full-color scanners.
And this is where things get interesting because fair use permits compies of material in the library for research. But if enough students scan journals at high resolution and then organize and exchange them through the Net, there will be an enormous levelling of the academic playing field. That is a time I look forward to with eager anticipation.
It only mentions paper as the object to take a picture of, but it might also work for objects further away. This could solve the problem of the often very narrow angle lenses those tiny cameras have.
Stitching multiple images automatically is nothing new but is CPU intensive. So Moore's law will take care of that.
Net sa best, mar it koe minder
I remember many, many years ago seeing people working on this sort of thing at the MIT Media Lab. The idea was that you could take a standard resolution video that panned across a scene, and by merging the frames over time create amazingly high resolution images. I remember motion being tricky to deal with (as in, things moving in the scene) because it would either confuse the algorithm that tried to figure out exactly where the camera was pointed for each frame, or cause things to blur. But if you panned across a landscape, the result was an amazingly high quality image.
Enable 3D printed prosthetics!
Well...
This has other uses... I've thought about it before. Like shooting panoramas. Stick the camera in the air... push the button and rotate.
Voila.. panorama!
Life is the leading cause of death in America.
Take care!
Erick
http://www.busyweather.com/
Wouldn't OCR be more difficult in Japanese than english? With english many letters are required to create a single word, thus if individual letters are not properly recognized they can still be determined by their context within both the word and the entire sentence.
In Japanese there are fewer symbols per word, many more symbols to choose from, and symbols that contain much more detail.
So I would think OCR in Japanese would be many times more difficult than OCR in english.
Finally, you now have a phone that is only useful for scanning Japanese. If it acted like a real scanner then it would be useful for any language.
Dan East
Better known as 318230.
There is one thing that is not clear yet: How far away is "over the surface"? I mean, looking at a piece of paper from 1 meter distance is "over the surface", but I doubt it will get a high-res picture on a cell-phone camera. If I get closer to the thing I want to scan, then the field of vision is getting smaller. At the end that means that a cell-phone camera laying on the piece of paper that it should scan will only "see" a very small part of the image. So if I have to move it along to get the whole image, I'll be busy for a while, the data stream will be quite big and if I'm unlucky the camera shadow will darken my scanned movie. If I scan from a distance of lets say 10 cm away then the question is how much influence a variation of this distance will have to the result. And how I know that I got all details of the picture. And when the camera memory is exceeded. :-)
This software is from the mid to late 90s and unfortunately not available anymore. iPIX purchased the company and discontinued all of its products. There are a few links to buy it but they say it's unavailable and I haven't ever been able to find it on file sharing.
Another interesting program they had is VideoBrush Panorama. It is can only stich vertical and horizontal pans (don't even try zig-zag). It's pretty cool to be able to get panoramas from video pans, and the software is very easy to use. There is no need for a tripod. You can get an evaluation copy here. This and a resource editor might come in handy if you want to use it.
With the problems that the DoD has had in the past with Cell Phones with Cameras I wonder if this will get them even more scared of such technology.
Imagine if they freaked out over 1Mega Pix cameras because they could take FUZZY pictures of classified docs - This kind of technology will send the DoD over the edge. As it is right now Cell Phones with cameras are prohibited in all classified environments (at least byt the NAVY that I know of).
A Cell Phone with this kind ouf tech could be banned from the ENTIRE base/post/shipyard etc. One of the things that the drill into your brain in the service is that over time a bunch of little bits of unclassigied data can be made into a very informative report that borders on the classified.
Just my 3MegaPix Worth
i do a bit of computer vision and here could be a basic method this this to work:
.. not that hard :)
for each image
fit an affine transform to the last
[this should work easily because
1) the paper is planar
2) the paper and it's background are hopefully different - with nice edges in between
3) the lighting conditions are the same (depending on how you hold the phone)
4) the paper is not moving
each of these tranforms can be applied cumulatively to the future images, though error is reduced by mapping everything to the center image.
this takes care of the registration problem (other techniques like KLT might be useful...maybe [ http://vision.stanford.edu/~birch/klt/ ] )
then you can apply techniques of super-resolution to get a higher resolution image [ http://www.ri.cmu.edu/projects/project_323.html ]
try it
having a rectangular, planar, still, evenly-lit piece of paper helps!
Robo-Blogs of the world: UNITE!
Its called "super-resolution" and there are a bunch of papers on it, using very different techniques, and different sources of images.
I also think this is used on Mars by the MOC team to produce 0.5m resolution images from 1.5m source data.
You can do this with a normal digicam btw, download registax 2 for example. Just take consecutive images of the same static subject, and combine them.
The key idea here is that several lo-res pix (recall that a camera phone can barely take vga still, and the movie modes will be even less impressive) can be recombined to make one hi-res picture.
You basically have to figure out where each lo-res picture goes and place it into the hi-res document. If you are careful, you can place several overlapping pictures with sub-(lo-res)-pixel accuracy, letting you increase the resolution even more. You use the fact that you have high accuracy in the time domain to help you out in the spatial domain.
This will probably be done out-of-camera.
This kind of thing would be excellent fun - and have some pretty neat practical uses too.
Until it becomes part of one of the iApps on a Mac I doubt it will be at all intuitive to use - so it wont be!
My dream application of this kind of thing? Pan around with a video camera semi randomly within a scene for an arbitrary period of time and have a bit of software capture this and 3D model the scene so you can walk about in there with very high resolutions. Any blind spots could later be refined with new video from a new location in the scene.
THATS why I'll be buying my 20GHz PVIII with a TB of RAM.
I don't see how this could *not* require a rediculously steady hand. I have enough trouble making my digital camera photos not blurred!
AS a visually impaired person born with a loke low resolution retina. I can say that I used this ton compensate my disability to see details if not near enough.
My brain compensated this by applying a continous eye movment (nystagmus). This allow my brain to get several low resolution moving pictures and be able to compute the missing sharpness and details.
Many born visually impaired have this nystagmus as some compensation.
I'am glad this become a mathematically and scientifically analyzed process. This is great it get some practical use. This remind me of the pictur analysis and filtering applyed to Hubble when it was known is main mirror could not focus correctly.
Léa Gris
ALE is an open source tool that does this nicely. It is normally intended for turning a large number of images of the same thing into one higher quality image, but when you use the --follow and --extend flags. it can turn a sequence of images from a video into a single larger image.
To quote from their site: ALE is a free software program that renders high-fidelity images of real scenes by aligning and combining many similar images from a camera or scanner. The correct similarity between images is roughly that achieved by a somewhat unsteady hand holding a camera.
http://www.chiltonwebb.com/iStill/
A lot of my friends who are in the Navy stationed out of Japan already have cell phones that can do OCR on them. Not exactly a scanner per se, but they can scan in text from a from, and considering that these cell phones also can interface with the new Memory Stick Pro (1GB), you can just go to the library/book store, and stand there and scan your brains away without buying the book. (Which is why a lot of book stores in Japan now have their books in shrink wrap to keep people from leeching/OCRing for free.)
The phones can also read barcodes, too.