Remember that MotionDSP stuff we talked about a few weeks ago? No? Let me refresh your memory: it’s a process — extremely CPU intensive, I’m thinking — by which video data is thoroughly analyzed and enhanced, reducing things like noise, stutters, color and contrast issues, and so on. TechCrunch mentioned it as being a web service, but now the MotionDSP guys are working with NVIDIA to port the processing engine to CUDA, which they say has resulted in major performance gains. Point being, this may be something you can do at home now instead of uploading it to be processed remotely in their clusters.

They say “real time” is now possible locally but I’m sure there’s a resolution or bitrate limit on that; this won’t be for clearing up glitches on 720p TV rips, but you might be able to use it to enhance some of those YouTube FLVs or jittery iSight videos. Two more demo vids after the hump.

The video above shows MotionDSP stabilization.

Below, frame interpolation (great when it works, but interpolation can go horribly wrong)

Lastly, cleaning up bad exposure and noise (noise not really visible at this res):

There are a lot of startups offering technological means of keeping their noses clean. Most of the solutions function as digital detectives, comparing the video fingerprints of copyrighted content with uploaded content for a match. Some of these companies include Audible Magic, Advestigo, Gracenote, MotionDSP, Philips, and iPharo. YouTube has implemented Audible Magic, although I haven’t noticed a difference. MySpace also incorporated Audible Magic but took the added step of banning re-uploading content violating copyright (“Take Down Stay Down” initiative).

However, while computers are great for solving well defined problems at a dizzying pace, they don’t always do that well when the rules become murkier. Judgments need to be made about whether playing a song or video constitutes “fair use” and simply changing a few characters of the title can fool more basic filters. That’s why 5-year-old BayTSP has decided to keep humans in the loop. The WSJ takes an in depth look at the company.

The Journal reports that BayTSP has hired more than 20 “Video Analysts” to watch videos and report copyrighted content starting at $11 an hour. Their searches are helped by BayTSP’s software, which most likely gives them a head start on what to look for. The company’s most notable client is Viacom, which it supplied with the data for their 100,000 video DMCA takedown request last year. Viacom says it pays BayTSP more than $100,000 each month for the service. The takedown requests have resulted in over 230,000 clips being removed from YouTube for Viacom. BayTSP says its error rate on Web videos is only around 0.1%.

Despite these efforts, video piracy remains rampant both on Google video search and many other social video sites. Once content is taken down, some users simply re-upload them to the site. MySpace is apparently countering this behavior through a file blacklist, but other video providers are certainly concerned with pushing away potentially valuable content and users. Content providers have continually leaned on the heavily manual DMCA safe harbor clause, while copyright holders clamor for embedded filtering. Google has recieved a long list of take down notices. AT&T has expressed an interest in filtering their network directly.

One thing’s for sure, there’s still a lot more debate needed amongst us humans before the computers chime in.

If you start seeing the quality of mobile shot video on the web improving, don’t assume it’s an improvement on the phone side. MostionDSP’s resolution enhancement could become an industry standard for video sharing sites. You can see the quality of the video enhancement on the company’s sample page or on the MotionDSP user page on YouTube. The difference isn’t huge, but when consumers had a choice between the two I think the demand will be clear.

MotionDSP began in 1998 as a US military funded project at UC Santa Cruz. Its first product provides real time enhancement (meaning a one-minute video will take one minute to enhance) using a system of 3 dual core servers. Those systems start at $30,000 but most large companies will want to buy them in bulk.

A company representative told me that future products could include chips to perform the resolution enhancement on phones or laptops. Anything that can run its algorithm is a potential platform, they said. They have taken about $500,000 in angel funding and are expecting to close Series A funding before end of year, probably in November.

While many of us try to take video and images with our cameraphones or webcams, the resulting footage is rarely usable and suffers from a great deal of anti-aliasing and generally messy pixels. MotionDSP has solved that problem by using special algorithms to “recreate” lost data in real time, thereby bringing almost useless low-light and compressed video into tight focus.

TechCrunch talked to the folks over at MotionDSP and discovered that the process works best on smooth, non-jerky video but, as we can see above, the improvements are immense. The product started as a military research project at UC Santa Cruz but is now available for licensing.

Startup Uses Military Tech to Fix Low Res Video [Techcrunch]