Cleaning Up Shaky Home Video
Video-enhancing software developed for the CIA is coming to consumers.
An annoying side effect of ubiquitous video cameras–on cell phones and digital still cameras, for instance–is ubiquitous low-quality videos. YouTube is rife with them: shaky soccer games, dimly lit baby’s first steps, and pixelated party clips. Next month, however, software maker MotionDSP, based in San Mateo, CA, will release a $40 download for PCs that could clean up some of the most annoying aspects of subpar video, from shakes to low light. The product, called vReveal, is aimed at the average home videographer who wants better vacation clips but doesn’t want to invest in serious video-editing software.
According to MotionDSP’s CEO, Sean Varah, the company developed its image-enhancement algorithms for the CIA. But the consumer version of the software comes with conventional video-editing tools that allow people to edit and stitch clips together, and it works with the Windows XP and Vista operating systems.
The idea of cleaning up video is certainly not new. For years, software companies have been using so-called super-resolution algorithms to make scenes clearer. In essence, these algorithms analyze information about the color and position of pixels within a handful of video frames and extrapolate extra information that can be used to make edges sharper or eliminate motion due to camera jitter. Salient Stills, an MIT spinoff founded in 2000, uses super-resolution algorithms and other techniques to enhance video for law-enforcement agencies. More recently, Intel has investigated the potential of running super-resolution algorithms on its multicore machines.
Apple, whose iMovie is generally considered the most useful consumer video-editing software, announced yesterday at the Macworld Expo that iMovie ‘09 will have some form of automatic video stabilization to clear up shakes. Additionally, Adobe and Sony offer editing software that includes video-enhancing features for under $100.
But Varah says that vReveal’s advantage is that it was designed specifically for the casual home user. “What [Sony’s and Adobe’s products] started off as were high-end video-editing programs,” he says. “The companies feature-stripped them to make cheap, under-$100 products that you can run at home.”
Varah also argues that having to meet the CIA’s requirements set a higher standard for MotionDSP’s algorithms. The algorithms leverage a handful of patents based on work done in the lab of Peyman Milanfar, MotionDSP co-founder and professor at the University of California in Santa Cruz.
The algorithms track motion through several frames of video to determine which characteristics of the motion appear to be due to a jerky camcorder. This motion is then subtracted from the scene when the video is rendered.
To increase brightness without diminishing picture quality, the software applies a similar trick. “Any software can bump up the brightness of a scene,” says Nikola Bozinovic, the company’s CTO. “It just scales the pixel intensity by a factor of two.” The problem is that this also amplifies the noise in the signal, which appears as graininess. MotionDSP’s approach eliminates noise by averaging it out over a number of frames.
Another feature of vReveal is that it can increase the effective number of frames in a video. Many cell phones capture video at a rate of 15 frames per second, but motion doesn’t look smooth to the eye until it’s close to 30 frames per second. By analyzing the motion of objects within multiple frames, the software is able to synthesize frames in between.
Laura Teodosio, cofounder of Salient Stills, notes that the sort of super-resolution technology that MotionDSP offers isn’t groundbreaking. “The thing about super-resolution is that algorithms and techniques really haven’t changed over the past couple of years,” she says.
But Teodosio is intrigued by the attempt to bring super-resolution tools to the consumer market. “I think the challenge for anyone in the consumer space is twofold,” she says. “It’s creating something that’s easy enough for consumers, and second, giving them enough change in their imagery that they’re going to come back and use it again and again.”
If vReveal is successful, MotionDSP’s algorithms could move into other types of software. By the middle of next year, the company plans to open up part of its source code to outside developers, says Varah. This means that a company like Skype could use MotionDSP’s algorithms in its video chat service, improving the quality of the video without requiring extra bandwidth. Additionally, Internet video providers such as YouTube and Hulu could take advantage of the software to bump up quality.
AI is here.
Own what happens next at EmTech Digital 2019.