Few of us give a moment's thought when a local TV news department fetches video clips of the mayor's speech from last year or a decade-old historic storm. Same thing when a reporter quickly inserts footage of a memorable touchdown run from several seasons ago or an arcane civic event pulled up from the archives.
Station asset management tools for such video searches have become embedded into standard editing equipment, including products from Avid, Front Porch Digital and Grass Valley.
Those relatively routine retrievals make use of various metadata capabilities that are now built into the production system. In particular, the labels and titles on clips in the station's video library or the closed-caption text on the report's narration make it simple to find the venerable images.
Now comes Web video with a far larger inventory of content--essentially endless--and far less sophisticated production techniques. A growing number of video search impresarios are promising to help "viewsers" (a neologism to describe interactive media consumers) find the video clips they want.
Google's $1.65 billion bet on YouTube (resulting in something known as "GooTube") brought the video search issues into greater focus, with observers expecting big moves based on Google's historic strength as an Internet search engine and its existing, albeit low-key, work on Google Video Search.
Not coincidentally, a slew of other video search initiatives is bubbling up. More than a dozen projects are exploring technologies ranging from pattern- and image-recognition to voice-to-text conversion. Startups such as Blinkx.tv, MetaCafe Inc. and PureVideo Networks are in the race, along with companies such as AOL and Kodak, as part of their re-invention processes.
Although much of today's video search activity is coming from Web-oriented enterprises, there is a simultaneous flurry of effort toward image recognition in the surveillance, defense and intelligence sectors. Eventually, some video search techniques and tactics will spin out from the government's heavy investment in those security technologies, which may be commercialized--although the timetable is not clear.
VIRTUAL TRIGGERS
For example, facial-recognition software, along with the tools to tag specific frames in a clip, powers much of the video search technology that is now being quietly deployed in security applications. It is a short hop from such police and business uses into entertainment and information search.
Here's how it might work: if a viewser likes a particular actor but does not know his name, the face could be used to trigger a search for other movies in which he appears. Although the possibilities for serendipitous errors are mind-boggling, the possibilities are also great for mining video libraries for old gems. On the news and information side, facial recognition opens the way to new kinds of digital reporting and local coverage.
Significantly, during the same week that the GooTube deal was unveiled, another video search pioneer, Blinkx.tv, allied with Microsoft, bringing Blinkx image and contextual search methods to the Microsoft Network and Live.com. That deal coincided with the Blinkx release of several enhancements to its image and contextual search methods.
"Video should be delivered visually, not as a text-based list of results," said Suranga Chandratillake, Blinkx founder and chief technology officer.
NEW MARKETS
Although Eastman Kodak Co. has not formally unveiled video search as part of its current product roster, the company's initiatives in still-picture pattern recognition could migrate into the video realm. Kodak CEO Antonio Perez has cited his firm's new "eFinder" tool as one of the products that Kodak will offer to help customers find photos and scenes from "digital shoeboxes" of old pictures.
eFinder is a pattern recognition technology that can be used with "intelligent archives," as Perez calls the image databases. Similar image recognition technology can be adapted to other still or video libraries. The next step would be picking a scene (let's say World War II battles) and then requesting a search for similar scenes--even if the viewser does not know the title of shows that contain such action. (Yes, genre lists of war movies might accomplish the same result in this case, but video search could deliver a far more focused roster of applicable options.)
This kind of contextual search is taking several forms, including ones that exploit the crossover capacities of the Web and existing software. For example, AOL and others are developing systems that automate the process of translating the audio track of a video program into text, and then using conventional text-searching tools to find a scene or clip.
More sophisticated contextual tools can handle video mapping and other factors to identify the location of scenes--Well over a dozen Web sites claim that they offer some type of video search today, with most of them focusing on Web video sites. They range from Yahoo Video Search, Google Video Search and AltaVista Video, to specialized video-sharing sites, such as The Open Video Project and Open Media Network. C-SPAN and PBS proclaim video search capabilities on their sites, although they seem to be, for now, mainly text searches for tagged presentations (e.g. C-SPAN's chronicles are tagged by the name of the member of Congress or the legislation/topic being debated.)
With the rise of Web video, the demand and capabilities of video search are destined to grow. Many of the tools developed for Web applications may find their way into broadcast asset management--or at least onto Web sites run by TV broadcasters.
