Sony Develops TVs That Zoom in for True Close-ups 275
prakslash writes "Sony has unveiled version 2 of its 'Digital Reality Creation' technology that allows viewers to pan around a TV image and then zoom in. Unlike the current TVs that simply scale the image, Sony's technology does 'true' zooming by digitally enhancing the signal to communicate gloss, depth and texture.
Digital Zoom is a MYTH! (Score:5, Informative)
Come on, it's trying to create data that just plain isn't coming from the original source, therefore it's nothing but guess and check logic. Sure it my smooth out what it thinks is a rough edge... but that's still only guessing and making up detail that just wasn't there.
digital zoom vs real zoom (Score:5, Informative)
unless tv has lower resolution than broadcast quality this is as fake as 200X DIGITAL ZOOM.
marketing hype (Score:2, Informative)
official press release (Score:5, Informative)
Re:Digital Zoom is a MYTH! (Score:5, Informative)
In fact, it may not only smooth out but actually create rough edges not in the original. Think of the case recently where the girl was kidnapped and the FBI enhanced the kidnapper's image from the crappy surveliance tape.
The article doesn't specify if it zooms a frozen screen (like a paused image) or during a sequence. Either way it could use past information from previous image frames to enhance the result.
Its one of those things that you need to actually see to believe the hype.
Just another algorithm..... (Score:5, Informative)
http://www.digitalanarchy.com/toolbox/toolbox_r
I'm guessing that Sony have simply come up with another one. Regardless of what they claim, you can't "zoom in" on an image with a fixed resolution, you're always going to be using some type of interpolation and this will introduce digital artefacts.
Good existing zoom implementations (Score:5, Informative)
Re:Sometimes, you don't want to see the gloss. (Score:3, Informative)
Translation: (Score:1, Informative)
It's a bird, it's a plane... (Score:5, Informative)
There's actually a whole host of algorithms that go well beyond the junk they throw at us for "digital zoom". The two most applicable algorithms for this particular problem -- increasing the resolution of video above and beyond the source data available in a particular frame -- are temporal integration (collecting data across multiple frames) and superresolution by example (automatically associating and recalling high resolution imagery when a low resolution equivalent is shown). Some example code:
Temporal Integration: ALE [dyndns.org]
Superresolution by Example: Image Analogies [nyu.edu] -- not automated, but remains one of the cooler pieces of code ever shown at SIGGRAPH.
From the article, I'm guessing it's another ALE style stacker. They probably needed to write one for their cameras anyway.
--Dan
Re:Digital Zoom is a MYTH! (Score:3, Informative)
I think I read about it in Popular Science or Popular Mechanics (Not sure though)
Re:yet another invention of the (not so ?) obvious (Score:5, Informative)
Yeah, its way harder. At least high quality image interpolation is. Theres been decades of research into it and dozens of different methods have been the topic of phd papers. Lots of high end math and very complex algorithms.
Ever printed a photo on an inkjet printer? You're seeing a pretty strenuous use of interpolation algorithms there. A typical resolution image coming off of a digital camera only prints at maybe 2 or 3 inches across at the resolution a typical printer operates. So if say, you want an 8x10, your printing software does some serious interpolating.
And not all printing software is equal, either. The algorithm makes all the difference. Its why you can get a so-so large image out of photoshop's print facilities (that uses bicubic) and a noticably better one from QImage (at the moment, pyramid)
Re:digital zoom vs real zoom (Score:3, Informative)
All consumer TV's have lower resolution than broadcast quality. If you buy a really expensive NTSC TV, it might be able to resolve 600 lines. The NTSC signal is comprised of 720 lines.
In HD, it's even more drastic. A really expensive HD set might resolve 800 lines of a 1,920-line picture.
A broadcast monitor that can resolve 1,000 lines costs $40,000.
Re:It's a bird, it's a plane... (Score:1, Informative)
Example-based superresolution is a little dodgier. What it does is construct a statistical model from high-resolution images of similar subjects. Then when presented with a new image, it uses this statistical model to "fill in the blanks" with statistically-plausible details. That probably wouldn't hold up as forensics in a court of law, because you're filling in the details with information not present in the actual image -- it comes from training images which are supposedly "similar to" what was present but not visible in the actual source. But it has its uses: suppose you train the algorithm with a bunch of text in various fonts. Then you can use it to zoom in on small and unclear text, and because alphabetic characters are all pretty much the same, it does a good job of reconstructing what the original letters probably looked like.
Re:It's a bird, it's a plane... (Score:1, Informative)
Re:Digital Zoom is a MYTH! (Score:5, Informative)
The basic idea is to use statistical methods over a series of slightly "jittered" video frames to create a high resolution or high quality still image. When an image in front of the camera is shifted by a non integer number of pixels, the images are lined up exactly with each other so that the edges of pixels overlap each other. Taking the average of these sup-pixel overlapped images at a higher resolution yields a higher quality image than the simple mosaic or blur you would get by scaling or interpolation. If the physical shape and response function of individual camera "pixels" is known, even more accuracy can be contained. The method can probably even be applied to rotating or enlarging/shrinking images of objects as well, but with more complex mathematical models for the motion and camera viewing transformation
A generic system as described in the article probably uses the frequency information about the image to construct the textures, but it wouldn't be difficult (but processor intensive) to track translational sub-pixel movement of objects and apply the above process to increase the resolution. MPEG already takes advantage of the fact that more compression for fast changes in an image are unlikely to be noticed, so it wouldn't have to improve the moving parts, just the 8x8 blocks that have B frames, since they are relatively unchanging. I bet they will even get a patent on the process, despite the fact that it's been published and I can think up most of the rest within a few minutes...
Re:Good existing zoom implementations (Score:3, Informative)
Re:digital zoom vs real zoom (Score:5, Informative)
Where did you ever get that idea? If that were true broadcast TVs and DVD would look exactly the same. I mean if broadcast TV is already better than a TV could support, how could DVDs possibly improve the picture? I would think anyone who has ever watched a DVD on their television has already empirically proved your statement incorrect.
In fact, broadcast TV is a far lower resolution than your TV can support (525 scanlines, of which 480 is picutre information, of which 330 is the theoretical max that will be displayed). Rather than try to explain it myself, some very good technical explanations of how it all works can be found here [aol.com] and here [doom9.org].
This sounds like a previously-disclosed technology (Score:1, Informative)
I don't think this is mere trollism on Sony's part, but just an implementation of what others have done.
True resolution of inkjet printers is much lower (Score:5, Informative)
Not true, because inkjet "resolutions" are really dot densities and not resolution (resolution would be how many distinct dots can you print per inch.) That's why laser printers with nominally "lower resolution" output crisper text. Also the dot density is for a single colour - complex hues such as skin tones have to be simulated by digital halftoning (essentially multiple dots forming larger colour pixels) techniques which reduce the effective resolution several fold depending on the colour being simulated and the accuracy desired. That's why continous tone printers such as dye subs with nominally "lower resolution" can give much sharper colour prints.
Software would have a major effect on the quality of colour prints from inkjets but that would mostly be from how the halftoning was done rather than the interpolation per se...
Digital TV is MPEG to start with... (Score:3, Informative)
This argument about "lossy compression" comes up again and again. You could say that everything is "compressed" from the original reality; it's a matter of whether you do stupid compression (drop the sampling rate) or intelligent compression (removing things that can't be heard/seen). It's all about getting the maximum perceived audo/picture quality with a given data rate. So, taking the same data rate, would you prefer "uncompressed" 8-bit video at 320x240 say, or MPEG4 "compressed" HDTV at 1920x1080?
Re:Digital Zoom is a MYTH! (Score:3, Informative)
This morning. Cut a couple of slices for my sandwich and one for toast. No big deal. Took me less than a minute for all three slices. (It helps if you've got a good bread knife.)