Tuesday, June 24, 2008

A Firm Makes Its Mark In the Blink of an Eye

FotoNation Team in Ireland
Solves Photography Pitfalls;
Next Up? Camera Phones
By WILLIAM M. BULKELEY
Galway, Ireland

This small city on Ireland's west coast has become the digital capital of the winning smile and the clear eye.

Two-thirds of the world's digital cameras -- and soon hundreds of millions of camera phones -- are outfitted with technology from a tiny hometown company, FotoNation, that has solved some of photography's perennial pitfalls.

Software developed by the company allows Nikon Corp.'s newest digital cameras to employ "smile detection" -- which stops the picture from being taken until everyone in the frame is grinning. It makes it possible for most Eastman Kodak Co. cameras to automatically scrub red eyes from photos as they are taken. And it enables the newest Pentax Corp. cameras to warn shutterbugs when a picture is spoiled because someone closed their eyes.

FotoNation was founded by its chief executive, Eran Steinberg, a 46-year old Israeli engineer who lives in San Francisco. Mr. Steinberg started working on digital-photography development in the 1990's with computer scientists at the National University of Ireland's Galway campus -- after he married an Irish woman. After his previous employer closed in the dot.com bust, he and the Galway engineers started taking on projects for camera makers.

Mr. Steinberg saw that cameras with a digital brain could fix problems that had bedeviled photography for decades. "Traditional camera companies look at improving image quality with glass," he says, referring to lenses. "We asked, 'How can you do things to the image?' "

FotoNation, which has 70 employees world-wide, divides its engineering efforts between Galway and low-cost Bucharest, Romania. At a recent visit to its offices in Galway, engineers from Ireland, Romania, Moldova, Germany, Hungary and India fine-tuned software. Some were writing new code for the picture processors in cameras. Others checked changes in the code against a database of 600,000 human faces that FotoNation has built to test its software algorithms.

Other companies, such as ArcSoft Inc., Fremont, Calif., and Omron Corp., of Kyoto, Japan, also market photo-improvement technology. But FotoNation has the largest share and the broadest product line, market researchers say.

"Those guys are geniuses," says the U.S. marketing chief for one Japanese camera company.

Tessera Technologies Inc., a San Jose, Calif., semiconductor company that specializes in products for cellphone cameras, completed its acquisition of FotoNation for as much as $39 million last month. FotoNation won't disclose its annual revenue.


FotoNation's first hit product addressed a problem that has vexed photographers since the advent of color flash-photography: red-eye. Red-eye occurs when the light from a flash penetrates a dilated pupil and reflects back the color of red blood vessels in the eye. Camera makers had tried various fixes, such as a preflash to contract pupils before the actual picture is taken, or separating the flash from the camera lens -- complicated solutions that drain batteries.

FotoNation's breakthrough idea was to develop a software algorithm small enough to run on the processor of a digital camera that can detect eyes in a picture, determine whether they are red, then automatically erase the color before the image was stored. Nikon's introduction of the feature in its CoolPix line in 2003 legitimized the idea of buying software from FotoNation and attracted other customers.

Last year, 80% of the 100 million digital cameras sold world-wide came installed with red-eye reduction. Most camera makers no longer give users the option of turning off the feature.

Mr. Steinberg says FotoNation hasn't yet been able to solve the unnatural coloration, often green in dogs, that shows up in many photos of pets. One of the biggest problems is that there is too much variation in dogs' faces to pinpoint the eyes.

Since the CoolPix, FotoNation has unveiled at least one new feature every year, mostly designed to improve photographs of the human face. "About 80% of consumer photos have faces in them," says Mr. Steinberg. If a camera detects a face, it can set exposure levels and distance on the face. It prevents a common mistake of snapping a picture of two people by aiming between them and inadvertently focusing on a distant object.


FotoNation's vice president of engineering, Petronel Bigioi, a Romanian who came to Ireland as a graduate student, says that FotoNation's face-detection algorithm was developed by having a computer digitally examine hundreds of thousands of images of faces and letting it figure out what they had in common. The computer created about 200 rules for detecting a face -- including rules for detecting two eyes and a nose, and other rules for measuring the distances among them. The FotoNation software decides it has found a face when it confirms about 20 of the rules.


One key to FotoNation's software is its ability to continuously examine the digital-image sensor in cameras before the picture is actually shot. Once it locks on a face, it keeps track of its location, in part because it knows faces don't move very fast. Face tracking isn't perfect. It will lose someone if they turn around and all it sees is their hair. FotoNation is still working on software to identify facial profiles -- which are daunting, because they show only one eye.

Last year, companies like Sony Corp. and Casio Inc., started selling cameras with FotoNation's face-identification technology. The idea is that photographers can teach the camera to recognize at least 20 different faces of individuals such as friends or family members. The camera will know to focus on them, instead of others in the picture. The feature, which Casio calls "Family First," can be turned off if the photographer intends to focus on one of the faces not stored in memory.

This year, FotoNation added a feature that Samsung Electronics Co. has included in some cameras called FaceTime, which tells the camera not to snap a photograph until someone has stepped into the frame with a smile.

FotoNation recently introduced software for eliminating "purple fringing," a distortion that occurs in areas of high-contrast shot with inexpensive lenses, especially when a picture is enlarged on a video monitor. Purple fringes may show up along a roof-line set against a bright sky, or around bare tree branches. Mr. Steinberg hopes the new product will be attractive to camera-phone makers.

Also in development: solutions to turn blurry shots caused by shaky hands into sharp pictures, and a way to enhance pictures taken in front of windows, whose brightness often trick cameras into turning people in front of the windows into dark silhouettes.

Mr. Steinberg says one reason he sold FotoNation to Tessera is to move the company's technology into a broader range of products. Tessera's chief executive, Bruce McWilliams, says there are potential uses in home security, medical monitoring and automobiles to warn drivers that blinking may mean drowsiness. "Human beings process life in video," he says. "This is what humans will want."

No comments: