Multitouch technology is receiving a lot of press these days. With Apple’s iPhone blazing the trail, devices that let users go from simply using one finger to press a button or launch a program to using two or more fingers to drag, rotate, and resize on-screen objects may soon send user interfaces that utilize a mouse or other pointing device the way of the dodo.

But a problem arises when the consumer’s desire for ever smaller devices comes into direct conflict with his desire for a cool multitouch screen interface. The smaller the screen, the more of it your fingers cover up during use. And because your fingers are finger-sized—not pixel-sized—it can be difficult to touch the screen precisely where a program is expecting input.

Researchers at Microsoft and MERL (Mitsubishi Electric Research Lab) have developed a possible solution: LucidTouch, a two-sided touchscreen device that uses the back of the screen for multitouch input while overlaying a semitransparent image of a user’s fingers so that he can see where he’s touching the screen as he interacts with a program.

“It’s often said that direct-touch is more ‘natural’ than indirect input devices,” explains MERL researcher Daniel Wigdor. “This may be true, but this naturalness comes at the high cost of occlusion and lack of precision.

Pseudo-transparency is a lightweight, intuitive way of recouping some of that cost while still maintaining the affordances of a direct-touch device.”

The team MacGyvered a prototype of the system by gluing a multitouch pad to the back of a commercial single-input touchscreen. Then, in order to capture an image of a user’s hands, they attached a Web cam to the back of the device using a long boom. Finally, software running on the computer to which the device was attached flipped the image and dropped out the background, creating the pseudo-transparent effect. In tests, users were able to type from behind on a split keyboard, browse a map to find a specific location, and drag and drop objects.

According to Microsoft researcher Patrick Baudisch, the next step in LucidTouch’s development will be to replace the external Web cam with an integrated video-capture device that can sense the location of a user’s fingers and create the pseudo-transparent effect. Some possibilities for achieving this include a capacitive array, LED sensors, or stereo cameras embedded in the device’s body.



Here Comes The Spider-Man . . . Suit

In last month’s “Under Development,” we joked that a new carbon nanocomposite material might be used to create X-ray glasses. It does seem that carbon nanotubes are being touted as the miracle material of the future, promising advances in chip technology, super-strong building materials, and clothing that can not only serve as body armor but also help humans climb walls, a la Spider-Man.

This time, the comic book reference is no joke. Nicola Pugno, a structural engineering professor at the Polytechnic University in Turin, Italy, has published a paper that outlines how carbon nanotubes could be used to create super-adhesive boots and gloves capable of supporting a human’s weight.

The inspiration for Pugno’s research came from the way that geckos and spiders are able to scamper up and down vertical surfaces, even very smooth surfaces such as glass, without slipping or sliding down. Geckos stick because they have thousands of tiny fibers on their feet that create adhesion through capillary action (due to a thin layer of water between their feet and the surface) and van der Waals force (a bond between close molecules). A spider climbing a web relies on a nano-interlocking of fibers.

In order to duplicate these adhesive effects on a human scale, Pugno theorizes that a material could be made from millions of carbon nanotubes woven into 1cm-thick threads. At the end of each thread, the nanotubes would fan out to provide millions of points of contact between the material and the surface, creating enough adhesive force to support the weight of a 150-pound man.

Although the suit is currently only a theory, Pugno hopes that a prototype could be developed from his calculations within 10 years.





Photographic Mix & Match

Even with digital photography, we still take plenty of snapshots that “would have been perfect, if only . . .” If only that building wasn’t in the way of the view. If only that obnoxious tourist wasn’t posing next to that statue. If only there wasn’t scaffolding covering the face of that beautiful cathedral. What if there was a way to convert those “if-onlies” into the perfect shots you intended?

Two researchers at Carnegie Mellon have developed a program that will help do just that. The software, called Scene Completion, uses a novel approach to image retouching. Unlike systems which attempt to reconstruct missing or removed image data extrapolating from other parts of the same photo, Scene Completion uses other photos—sometimes completely unrelated to the subject matter of the photo being retouched—to find an appropriate and believable patch.

In order to accomplish this, the software sorts through millions of photos (in this case, using over 2 million Flickr photos) as its data set and applies a “gist descriptor” algorithm (developed at MIT) to find photos that share similar general properties of the photo needing work: shapes, colors, geographical features, and textures. From that subset, the software looks for specific pieces to fill in the needed data, trying to match the colors and border well enough that with blending of the edges, the photos look almost as good as the real thing.

The program does have limitations. Some photos are more difficult to match than others, explains James Hays, the Carnegie Mellon computer science graduate student who developed the software. “Scenes that are very rare (an unusual focal length, orientation, camera height, subject matter, coloring, etc.) have fewer good scene matches. It’s harder to fill holes in those rare scenes since it’s harder to find similar scenes to take content from.”

The other major issue is the large number of photos needed to create an accurate subset from which to cull the most realistic photo patch, and the potential copyright violations if public photo repositories are used. Hays believes that using only the 75 million photos on Flickr, which are licensed to be modifiable under Creative Commons, provides a large enough sample size, but the question of how many of those images are of high enough quality to be useful remains.

There are currently no plans to commercialize the Scene Completion system, though the team would like to release it to the public once some speed issues have been addressed, perhaps as an online version. Currently the database size—half a terabyte of images—prohibits distribution in any other manner.


Memory Goes 3D

The next major advance in memory chips could come by working in the third dimension. Stuart Parkin and his colleagues in the IBM Almaden Research Center’s (www.almaden
.ibm.com) SpinAps group are working to develop a new type of memory that could store significantly more data in the same physical space (as well as access that data at much faster speeds) than current magnetic and solid-state memory devices.

The potentially revolutionary technology, dubbed “racetrack memory,” utilizes millions of nanowire loops positioned vertically around the edges of a silicon chip, each of which could store between 10 and 100 times more data than today’s flash memory. Electric current moves tiny magnets up and down the wires at speeds over 100 meters per second, allowing read/write speeds as fast as a nanosecond. This would overcome a major drawback of flash memory—its slow write speed.

The memory is in the early stages of development: Parkin has yet to build a prototype, although he has shown that the basic elements of this new type of memory are possible. Parkin’s team still needs to reduce the amount of current necessary to move the information along the nanowire and research the possible interference between the closely-spaced nanowires to determine how densely they can be packed. If Parkin is able to solve these problems, it’s possible that racetrack memory could make its move on the market in three to five years.