Pages: pp. 20-22
Rice University researchers have developed a hardware-software approach that promises smaller, less expensive digital cameras that use batteries more efficiently. This would enable digital cameras to work longer and eliminate the need to change batteries often, long a problem for users.
Generally, digital cameras collect all the information from a target scene and compress much of the data using a compression technology, typically JPEG, to reduce the amount of processing necessary.
"When it works correctly, compression removes redundant info, and so the images remain more or less the same visually," explained Rice University researcher and professor Rich Baraniuk.
Figure Rice University scientists have developed a new approach that promises less costly, less bulky, and more energy efficient digital cameras. Standard cameras use millions of sensors to gather information about a scene and then compress the data to reduce the processing load. The Rice technique employs one sensor and uses algorithms to extrapolate the small amount of data into a full picture.
However, he added, "Compression consumes quite a bit of battery power in a digital camera."
The Rice technique, on the other hand, would collect just enough information from a target to extrapolate into a full picture, without requiring compression.
To begin the process, algorithms in an external computer determine whether an image can yield a good picture via the new technique, Baraniuk explained.
The Rice research used an emerging field of information theory called compressive sensing, in which algorithms take a few randomized measurements of a scene and extract information for the whole picture.
The prototype device uses one photo diode sensor, so in effect, it's a single-pixel camera, Baraniuk noted. Conventional digital cameras contain millions of sensors, each of which represents a pixel.
In their prototype camera, the Rice researchers used a one-square-inch array with 786,000 tiny mirrors, based on Texas Instruments' digital micromirror technology, already utilized in some digital cameras.
The camera randomly turns half the mirrors on and the other half off, about 200,000 times per photo. The mirrors direct the light to a second lens, which collects it and focuses it onto the sensor.
The data gathered for the photo is based on the average light intensity of the "on" pixels, as measured by the sensor. The camera's algorithms then extrapolate the data to build the entire photo.
Eliminating compression lessens the drain on cameras' batteries, and decreasing the number of sensors reduces a camera's size and cost. Using the Rice technique with conventional digital cameras that offer relatively few megapixels could boost their image resolution.
This kind of camera could also be used for nonconsumer purposes such as infrared imaging and terahertz imaging, commonly employed in security and spectroscopy applications. Current digital cameras can't work with the light waves that these applications use.
The Rice scientists still must do considerable work before their camera could be commercialized. For example, the prototype device is about the size of a tabletop and its image-extrapolation algorithms are slow.
Baraniuk said his team is also trying to make the array's mirrors flip faster, to accelerate the image-acquisition process.
"It is an interesting, potentially useful technology for consumers, although it's too early to say whether it will ever appear in digital cameras," said Harry Wang, a digital-consumer-imaging analyst with Parks Associates, a market research firm. And, he added, finding a first adopter may be a problem because manufacturers want to recoup their investments in their current technologies.
In response to requests for comment on the Rice research, Kodak declined and other major camera manufacturers didn't respond.
Researchers have developed a new technique designed to protect online advertisers from click fraud.
Many companies pay Web site owners who host their advertisements every time that someone clicks on one of their ads. However, there have been cases of fraud in which, for example, Web site owners have hired people or used bots to click repeatedly on an ad to drive up their revenue, or other advertisers have clicked on ads to exhaust their competitors' budgets.
Indiana University and RSA Laboratories, the research arm of vendor RSA Security, are developing the Premium Clicks technique, which works by identifying clicks likely to be from actual shoppers.
This contrasts with standard approaches that try to identify only fraudulent clicks, explained Indiana University associate professor Markus Jakobsson. These techniques miss many fraudulent clicks, he noted.
The standard approaches work with Bayesian heuristic filters that use observations of previous behaviors to develop rules for determining the probability that a click is fraudulent. For example, one rule might state that if many clicks originate from an identical IP address within a short period of time, they are probably fraudulent.
However, there are numerous ways to commit click fraud, including many with no telltale signs, which make it difficult to create an effective automatic-classification method, Jakobsson said.
With Premium Clicks, if users visit a trusted Web site and conduct a qualifying transaction such as a purchase, the site will cache a cryptographic token—called a cache cookie—on their browser. If the users then click on an advertisement at another site, the system releases the token to that second site, thereby identifying them as a legitimate visitor who does business online, not someone committing click fraud.
Unlike traditional cookies, the cache cookies are kept in an area of the browser associated with storing images. In this area, the cache cookies are harder to remove.
According to Jakobsson, the cookies maintain user privacy via anonymous authentication. They log only general information about a purchase, enough for the system to authenticate the user but leave out specific data such as the buyer's name, he explained.
Premium Clicks might not identify all legitimate activity, he noted, but almost all of what it identifies is legitimate.
Nonetheless, click fraud will persist, so companies must continue monitoring their online advertisements, said Forrester Research senior analyst Shar Van Boskirk.
RavenWhite has released a Beta version of the technology to selected parties and plans to license the technology to vendors for commercial release later this year.
Many aspects of computer technology have changed and improved over the years, but one crucial element has not: the quartz-based clock that paces the work done by traditional synchronous processors.
Now, SiTime, an integrated-circuit designer and vendor, is about to release a silicon-based oscillating clock that would be smaller and less expensive than quartz clocks. This, in turn, could enable the production of smaller and less costly devices such as portable media players and digital cameras.
A traditional chip's clock is an oscillating quartz crystal that vibrates at a regular frequency, measured in gigahertz or megahertz, upon receiving an electric current. The clock synchronizes every action of a chip, controlling the order in which the processor performs necessary tasks, including sending signals to circuits and regulating the data flow.
During the past 40 years, quartz technology has advanced somewhat, but the fabrication and packaging has changed very little, so the clocks have not gotten significantly smaller or less expensive to produce, said Joe Brown, SiTime's vice president for strategic alliances.
Figure A company has developed a new silicon-based clock to be used as a smaller, less expensive version of the quartz clocks that pace the work done by traditional synchronous processors.
Silicon-based clocks are less expensive to make because manufacturers can use well-established silicon mass-fabrication processes, Brown said. "Silicon will yield tens of thousands resonators from one silicon wafer," he noted. And, he added, as technology improvements enable smaller clocks, the yield per wafer will rise and the cost will drop.
Quartz, on the other hand, must be carefully polished to specific dimensions to achieve the desired oscillating frequencies for clocks.
And then, said analyst Carl Howe with market research firm Blackfriars Communications, quartz clocks must be placed on a circuit board, which adds cost. Silicon clocks can be etched right into an integrated circuit like any other piece of a chip, he noted.
Also, the clock mechanism can be built smaller in silicon. SiTime is making a silicon clock that is just 300 micrometers square, while an equivalent quartz clock is about a millimeter wide.
SiTime's microelectromechanical system (MEMS) clock uses electrostatic forces to make multiple pieces of silicon resonate at a given frequency.
Users could feed the given frequency of one silicon clock into a phase-lock loop electronic system that multiplies or divides the frequency, yielding different levels. A quartz clock produces just one frequency.
Researchers at various universities have worked on silicon clocks for decades, Brown said. However, he explained, they've been unable to develop clocks as consistently accurate as those made of quartz because the silicon versions were sensitive to temperature changes that caused unstable performance.
However, improvements in MEMS fabrication—specifically the use of high manufacturing temperatures to remove silicon contaminants that cause temperature sensitivity—have made high-quality silicon oscillators more practical.
"Silicon clocks traditionally have not been as accurate as quartz. This is one of the problems that silicon-clock companies claim to have solved," said Howe.
The first silicon clocks could replace quartz clocks in simpler applications, such as basic microprocessors, but not in complex chips such as those used in GPS devices, which require high precision levels, according to Brown.
"The big challenge for silicon-clock adoption" he said, "is that quartz has been used successfully for over 40 years."
A research team has developed a computer program that wins at Scrabble not only by maximizing its own score but also by minimizing opponents' scores via analysis of their moves.
Assistant professor Eyal Amir and doctoral candidate Mark Richards at the University of Illinois at Urbana-Champaign (UIUC), developed InferencePlayer as an application that works with and plays against the open source Quackle Scrabble program.
InferencePlayer has not competed against any elite human players in tournaments yet but has beaten Quackle, Richards noted.
Basically, InferencePlayer uses information about the tiles its opponents most recently played to infer what tiles they might have and what their next move will probably be, enabling the program to block their ability to form a high-scoring word.
For example, Amir said, if an opponent plays the word "pigeon," the tile left in their rack probably isn't an "s" because if it was, they likely would have played the word "pigeons" to get the 50-point bonus for playing all seven of their tiles at one time.
InferencePlayer's algorithms consider all the possible letter combinations opponents might have held when they last played and determines their relative likelihood, he explained. The program then focuses on the most probable ones.
Tracking tile plays is what elite and competitive Scrabble players typically do, Amir noted.
This ability to precisely infer information from play to minimize opponents' scores differentiates this program's strategy from that of other Scrabble programs—such as Quackle, Maven, and Crosswise. They store entire dictionaries so that they can pick their own best moves and maximize their own score. InferencePlayer also uses this approach.
The UIUC researchers say they may try to improve their program's ability to create a detailed model of an opponent's play.
Richards noted, "We've been in contact with the authors of Quackle and are planning to work with them to include our player in their upcoming version."
Also, he said, "We believe our study of Scrabble can help us improve computer play in other games, such as poker and its variants."