LTU Makes Sense of Visual Content
Posted Aug 11, 2003

Though there are many tools available for searching through tons of text, tools for helping people search through terabytes of digital images are scarce. No wonder. Image recognition is a daunting task. There are currently more than two billion images on the Internet alone, according to estimates. To a computer, however, an image is nothing more than an array of numbers that describe pixels. Consequently, image recognition-based computer management solutions have remained relatively simplistic, often relying on manual intervention. For a long time now, image recognition and interpretation has been a job that only the human brain could handle.

With the explosion of digital content, there is an enormous need for an effective visual asset management technology that can really understand a digital image (or digital video), something that can actually analyze the content of a digital image and infer what the image is. Is it an apple? A logo? Pornography? A photo of a baby?

LTU Technologies thinks it has the technology that fits the bill. LTU says its software tools can "make sense of visual content." The company has developed patented image recognition technologies to automatically analyze and describe the content of emails, Web pages, streamed video, and digital video files.

Based in Paris, France, LTU Technologies recently opened a U.S. office in Burlingame, California, and has set its sites on the American market. That market is a potentially lucrative one, according to Jody Webster, general manager of the North American operation.

"The exponential growth of the Web has resulted in a big problem that impacts everyone who uses the Internet: digital information, especially visual content, remains difficult to access, protect, and use effectively," says Webster. "We believe that an efficient technology that can help locate, index, and track a specific image, or other visual content, has enormous potential in today's information-rich society."

She also feels the market is wide open. While there have been various attempts to introduce visual content management tools, many of the technologies in use today are text-based and overlook the essential need to manage content by understanding images themselves, she says.

Four Key Markets

LTU has identified four keys markets for the various flavors of its image recognition software.

First is content filtering. LTU Technologies' Image-Filter tool gives users a way to control the viewing and distribution of adult or objectionable visual content. A content filtering solution must be accurate enough to catch even a slightly offensive image, yet allow for local control of the flow (and blocking) of digital information based on user-specified sensitivity levels, says J. J. Wallia, director of sales and business development for LTU's North American operations. LTU's approach examines and "scores" the visual content of email, video, and Web sites so that parents, public libraries, and organizations can maintain local control of what passes through unblocked.

According to Wallia, his company's Image-Filter software is 89 percent-effective at identifying and filtering out pornographic images. He explains that the software can be "trained" to recognize certain shapes that are common in pornographic images. In fact, it "recognizes" 50 different pornographic shapes, according to Wallia. He also reports that the software has been trained to recognize and filter out other offensive shapes and symbols such as the Nazi swastika.

LTU's porn filtering technology is currently being used to help U.S. Customs create a "child exploitation database," according to Wallia. It has also been successful employed by the French police. Wallia points to a particular case in which the French police used LTU software to nail a perpetrator. A man was picked up on suspicion of creating child pornography, but his image was not in any of the confiscated pornographic photos, so the authorities had no concrete proof linking him to the creation of the images in his possession. But most pornographers shoot photos in sequential groups, using similar backgrounds, so the police "instructed" LTU's Image-Filter to search through the police's pornographic images database to look for images similar to the ones confiscated from the suspect. And, says Wallia, "Lo and behold, there in one of the other photos was the suspect." He's now serving hard time.

The second key market is image indexing. Because unorganized content has little value, many technologies and business solutions have been developed to help manage the enormous volume of text, images, and video that exists on the Web. Yet few solutions available today can facilitate the laborious task of indexing and cataloguing visual content. LTU's Image-Indexer translates visual content into semantic descriptions to enable accurate indexing. For example, the software can be used to automatically suggest descriptive annotations for digital images that have been uploaded to the system. Possible applications of the technology include both corporate solutions (the cataloguing and managing of a company's internal visual assets, for example) and consumer solutions (the automatic indexing of family snapshots, for example).

The third key application is searching image databases. LTU's Image-Seeker technology promises to help people make better use of massive data repositories and to develop the means to generate revenue from them. For example, online retailers have long been looking for a way to provide shoppers with an intuitive way to search and navigate an online store. With LTU's Image-Shopper software, retailers can implement features such as "find me a brooch that looks like this," or "show me more pants like that." This technology also unlocks intriguing possibilities for cross-selling or up-selling premium merchandise.

The fourth category is digital rights management. Because images are powerful, they are often of strategic importance to businesses. Corporate legal departments, software publishers, service firms, and patent offices are dedicated to protecting visual assets, from logos to industrial designs, from original artwork to new inventions. LTU's Image-Watcher technology can assist corporations in the tracking of unauthorized distribution, posting, use, or infringement of valuable proprietary assets such as photos, brand identities, and product designs—whether inside or outside the company, via emails or over the Internet.

One digital rights enforcement company that is using Image-Watcher is NameProtect. Using Image-Watch and an intelligent crawler, NameProtect continuously scans the Internet on behalf of clients like Pepsi, checking to make sure no one is using the Pepsi name or logo in an inappropriate manner. Another NameProtect client is Reuters, the international news agency. Some small Internet news organizations and certain individuals have made a habit of grabbing Reuters news photos and publishing or using them without crediting or compensating Reuters. With the help of NameProtect and Image-Watcher, Reuters is cracking down on photo thieves.

The Video Connection

While it's a bit harder to search through moving images thanit is through still images, the same technology still applies—there's just a lot more images, says LTU's J.J. Wallia. His company's product in this area is Video-Indexer, a tool that has recently found successful application in the hands of French TV broadcasters.

M6, one of France's top TV networks, is currently using Video-Indexer on its "M6 Web" Web site (www.M6.fr) to automatically index, cut, and publish online segments from "Six Minutes," its daily TV news program. By repurposing its footage, M6 hopes to maximize the value of its TV content while providing M6 Internet users with a better way to access relevant and structured information in real time.

"We needed a robust solution to automate TV-to-Web publishing. Since the implementation of LTU's technology, M6 Internet users access more rapidly and efficiency the specific content they are looking for. Using Video-Indexer makes M6 users' loyalty stronger than ever," says Alain Fortune, CTO of M6 Web.

Besides enhancing the Internet experience for surfers, the LTU software is also making M6 broadcast technicians happy by streamlining the TV-to-Web production process. J. J. Wallia says that in the days before M6 began using Video-Indexer, it would take a staff video editor 30 to 40 minutes to edit clips out of the "Six Minutes" show for Web use. Now, with Video-Indexer, most of the work is done automatically, and editing time has been reduced to less than five minutes.

Video-Indexer is able to accomplish this because its image recognition engine is able to instantly recognize the "in" and "out" points of the news segments. This is relatively easy, in the case of "Six Minutes," because the show's segments are clearly marked by graphic screens with titles such as "Business News" or "Sports." The M6 staff editor simply trains the software to recognize these standard segment title screens. Thus, much of the work of editing and assembling clips from the broadcast show for use on the M6 Website is automated.

--Mark Fritz