WASHINGTON — The CIA is bankrolling efforts to improve technology designed to scour millions of digital photos or video clips for particular cars or street signs or even, some day, human faces.

The innovative software from fledgling PiXlogic LLC (search) of Los Altos, Calif., promises to help analysts make better use of the CIA's enormous electronic archives. Analysts also could be alerted whenever a helicopter or other targeted item appeared in a live video broadcast.

PiXlogic plans to announce Wednesday that the CIA's venture-capital organization, In-Q-Tel, has invested an unspecified amount to help the company improve the software.

In-Q-Tel -- named for "Q," the fictional inventor of fanciful spy gadgetry for James Bond -- makes about a dozen such investments annually with roughly $35 million it receives from the CIA's Directorate of Science and Technology (search). In-Q-Tel was created in February 1999 and has gained favorable reviews from Capitol Hill.

"There was a great deal of interest in these capabilities," Mike Griffin, In-Q-Tel's president said of the picture-monitoring effort. "Because more and more of what is on the Internet is in visual form, the ability to search on those materials is important and getting more important all the time."

Neither In-Q-Tel (search) nor PiXlogic would disclose terms of the deal, though executives said similar arrangements typically have been between $500,000 to $2.5 million. In-Q-Tel is organized as a nonprofit firm, and Griffin said it doesn't put onerous conditions on companies that earn its funding.

"We're making a bet. We want them to succeed," he said. "This is a way for government to tap into cutting-edge, state-of-the-art technology."

PiXlogic's founders include Vladimir Troyanker, a Russian who left the company shortly after obtaining a U.S. patent for early generations of the search technology. Troyanker said he never dreamed when he moved to the United States in 1998 that the CIA -- his native country's arch enemy during the Cold War -- might find the technology useful.

"The world then and the world now are two different places," said Troyanker, who works as a contractor for a telecommunications company.

PiXlogic's chief executive, Joseph Santucci, said the company is probably one year from adding effective, face-recognition features to its software to help the CIA track photographs or videos of specific people.

The cloak-and-dagger software presently has only rudimentary ability to recognize a person's face and then find other photos and video footage of that person, although a demonstration of this feature by Santucci worked almost flawlessly.

PiXlogic executives hope to work through other companies to bring a version of the technology to consumers who have digital cameras and need help organizing their libraries.

Many current visual-search products rely on photographers or editors to manually assign keywords, dates or categories to photographs or videos describing their content, such as "soccer game" or "President Bush speech." A few systems can study prominent colors in a picture to recognize that objects, for example, are mostly yellow or blue.

PiXlogic's software analyzes each photograph or video frame, identifies items by geometry, color and other qualities, and stores those details in tiny computer files associated with each image. It can quickly compare details from a sample image -- a photograph of a type of car, for example -- against details from millions of other images in a private picture library or on the Web.

"We're able to emulate the sight process pretty well, but there are no cognitive capabilities built into the software," Santucci said. It can recognize the side view of an automobile, for example, but can't infer how a front or rear view of the same vehicle might look.

Some photography experts expressed doubt that an automated system could yet categorize and find images as well as humans.

"That stuff has not been widely adapted so far," said John Harrison, a vice president at MerlinOne Inc. (search) of Quincy, Mass., which makes photo-retrieval software. "It's not that the technology or concept isn't interesting, but its current level of development isn't quite there."

Harrison said fool-proof face recognition is particularly difficult for computers.

"It's easy to disguise yourself to fool that kind of software," he said.