Hitachi makes like Toshiba and touts "super resolution technology"

We actually had to dust off our spectacles to make sure this wasn't some sort of misunderstanding, but clearly, Hitachi is pulling a Toshiba. Generally, we wouldn't make such a claim, but given that Tosh also calls its wizard-like upconverting technology Super Resolution, we'd say the likeness requires such a comparison. Reportedly, Hitachi has been toiling day and night on its own "super resolution technology," which will theoretically take SD images and scale them to near-1080p. Amazingly, we're told that it can also "convert HDTV images, such as digital terrestrial broadcasts, into images with even higher (what, SHV?) definition." The tech is executed on a frame-by-frame basis and should be exhibited at the upcoming CEATEC trade show in Japan; as for a release, Hitachi's currently aiming to get this into LCDs and plasmas "after 2010."

[Shinigami @ Sep 25th 2008 10:18AM]

Just a reminder:
"After 2010" we'll have OLED HDTV mass produced so LCD won't matter any more.

[SHoe @ Sep 25th 2008 12:43PM]

just a reminder:
the upscaling technology in question has nothing to do with the actual screen display technology - so LCD / OLED is all the same as far as upscaling / not upscaling / Blue-Ray / whatever. This is pure image data manipulation.

[Shinigami @ Sep 25th 2008 1:24PM]

I totally agree but
"Hitachi's currently aiming to get this into LCDs and plasmas"
No word on OLEDs getting this :)

Besides I'm somehow certain nobody will want to buy a new SD TV any time soon... Will it be able to upscale HD-Ready to Full-HD and make it look even better?

[anantha92 @ Sep 25th 2008 10:31AM]

Wow, If that actually works as well as the preview, then thats amazing!

[Jeebus @ Sep 25th 2008 12:27PM]

It won't.

The image on the left is downres'ed from the image on the right.

[Jason @ Sep 25th 2008 10:59AM]

You can't add detail that wasn't in the signal to begin with. Apparently this is just better sharpening, or something like that.

A better solution is to sell TVs that actually match the resolution of the shows coming down the pipe! Eventually all 1080p TVs will actually have 1080 horizontal lines.

[Vidikron @ Sep 25th 2008 1:41PM]

You can fake the detail though by making a good "guess"... which is basically what upscaling does. I'm not a fan of upscaling video for that reason and I've never understood why people make such a big deal over it. To me its analogous to optical vs digital zoom on cameras. Most people hate using the digital zoom, yet many of the same crowd will rave over high-end upscalers.... it's all fake in either case.

[xcrunk @ Sep 25th 2008 1:42PM]

Your source image is 640x480(DVD) pixels which equals 307,200 pixels total.

1080P(Bluray) is 1920x1080 which equals 2,073,600 pixels.

Do the math. Upscaling is for morons who can't tell the difference.

[Chris @ Sep 25th 2008 11:38AM]

I can see artifacts in the super up-conversion screenshot alone. None of these technologies come close to true 1080p HD.

[konshuss @ Sep 25th 2008 8:12PM]

duh

[Rick @ Sep 25th 2008 7:29PM]

Cranking up sharpness adds so much unseen detail. That's why Best Buy does it!

;)

[Ysleiro @ Sep 25th 2008 11:52AM]

I think it's a good thing to invest in new tech. I can't blame Hitachi for wanting a piece of the HD pie.If they can truly back up that claim and deliver > 720p out of

[mb @ Sep 25th 2008 11:59AM]

looks like all it is is a realtime unsharp mask/edge sharpen....

[vagvoba @ Sep 25th 2008 12:33PM]

"Super resolution" is not a simple spatial filtering, like sharpening and the likes.

Super resolution algorithms take advantage of the temporal information as well. This is how it works:
- Let's assume that our spatial signal (the image) is under sampled. It means that the optics that was used to capture the image projected higher spatial resolution (sharper) images to the detector (CCD) than the resolution of the detector itself. In this case, if we zoom into the image to see the individual pixels, we will see aliasing artifacts because of the under sampling.
- Aliasing artifact is in fact the higher spatial frequencies blended into the lower spatial frequency signal. It means that the higher frequency info is actually in the image, but we cannot separate it and unfold it to make a higher resolution image because this info is blended into the lower frequencies.
- Now the trick of Super resolution is, that we try to track the image motions with sub-pixel accuracy on multiple video frames and using these frames, we try to iteratively separate the aliasing from the low frequency signal.
- In optimal case, we can reconstruct much of the high frequency signal from a dozen frames. The more frames we have, the more info we can reconstruct.
- Of course the whole thing works only of the camera or the objects on the image moves a little. It doesn't need to move much. Even small vibrations, sub-pixel motions are enough for reconstruction.

All in all, it is in fact possible to create higher resolution images from a series of low resolution images.
The whole thing is feasible but requires an enormous computation power to do it.

[SHoe @ Sep 25th 2008 12:58PM]

@ vagvoba
Yes indeedy - a lot of processing power.

It also isn't perfect, as even with a great deal of processing power, it is impossible to have perfect enough object motion recognition, such that it can say 'this object in frame B is the same object in frame A but it has moved this % of the screen to the right, and rotated on this axis this much and this axis this much (3D) - so now let's adjust for that motion, analyze the picture elements and combine the sampled elements from multiple frames to render a more detailed image'. There is still 'unknowable' detail that might won't be revealed by motion and the algorithms for (or AI might be a better term with the way this stuff is going!) object recognition and analysis can never be perfect enough to compensate accurately for the motion that is occurring in 3 dimensions.

But it is damned ingenious and I can see how they will be able to get CLOSE to actual 'real' HD. Still won't be the same though.

[mb @ Sep 25th 2008 1:15PM]

t vagvoba:

Very interesting, didn't think about temporal analysis.

However the article says: "Hitachi's super resolution technology is executed on a frame-by-frame basis. The technology is not designed to be applied to multiple frames..." So, I'm not sure you're on the right track, and the amount of processing power required for what you describe seems to be something that might be a little much for an embedded system.

[vagvoba @ Sep 25th 2008 1:37PM]

I just read the linked article.
The Hitachi guy who talked about the technology is a real dumb ass. Must be a marketing manager because he had no idea about what the algorithm actually does. However that is clear from his pseudo-tech-gibberish that it is not a real "Super resolution" algorithm.
In fact someone should sue them for their lies about the technology used.

[XenoX101 @ Sep 27th 2008 12:05PM]

Wow, thanks vagvoba, that was very well said.

[bandigolo @ Sep 25th 2008 12:38PM]

uh, photoshop --> sharpen --> more sharpen.

I hate to judge the entire technology from one iffy screen, but seriously, that looks terrible.

When are people going to realize that you can't generate resolution that never existed in the first place? It's just not possible. What is possible is by 2010 everything will be HD and this will be irrelevant.

[XenoX101 @ Sep 27th 2008 12:09PM]

I very much doubt in 1.2 years we will have HD quality video coming from our mobile phones and compact digital cameras.

[bandigolo @ Sep 29th 2008 6:05PM]

wtf? What are you talking about? This is a television. I'm talking about television signals - i.e. Blu-Ray and HD broadcasts. Who the hell watches their phone on a television?

[Ian Kennedy @ Sep 25th 2008 1:09PM]

Temporal-spatial sharpening is not new. This would fall flat on static images.You can only syntehsize so much detail from nothingness and I bet this resutls in weird artifacts with real-world content.

Get over it guys, Sony won. You lost.

Nobody who actually knows what this does cares about SD. This feature is the first thing anyone who gives to shits about video will disable.

[tommy @ Sep 25th 2008 1:09PM]

This tech does no better than regular upcomversion, my eyes can tell no difference. Toshiba is foolish for believing that this will replace Blu, shame on you toshiba!

[michael @ Sep 25th 2008 6:14PM]

your eyes must really suck because this article is about hitachi, not toshiba.

[Shinigami @ Sep 25th 2008 1:28PM]

I wonder...
If you can buy an upscaling TV or a Full-HD TV in 2 years... what will you get?
If great upscaling comes as a bonus, I'll go for upscaling Full-HD. If its "either or" choice, I'll go for Full-HD.

Doesn't really matter at this point - 2010 is still faaaaaaaar.

[michael @ Sep 25th 2008 6:17PM]

Shinigami....you're an idiot. all hdtv's upscale sd images to the resolution of the tv, be it 720p or 1080p. super resolution upscaling merely upscales the image to look less blurry than regular upconversion.

[Winter007 @ Sep 26th 2008 4:00AM]

You acctually can get more information from the picture, than the picture holds. Its quite a brilliant thing and people make nice $ on it. Ussually old movies get saved with it. Try googling it :)

They just took the old fact that "a group is worth more than the sum of all of its parts"...
The "extra" information is hidden in the fact that u get 25-30 frames per second in a movie picture. Clever software can to some level
apply information from more frames together, into a "richer" image (here a HD image). Ofcourse this works best with alot of movement in the picture, for instance, a still picture ( no changes in the frames) can not get any better, as u dont gain info with more frames...

getting to deep...

Google it :)