High Def Forum
Thank you for visiting. This is our website archive. Please visit our main website by clicking the logo above.

Upscaled DVD movie vs HD-DVD movie

Troymouni
04-18-2006, 10:14 AM
Theoretically speaking,wich one would look better?An upscaled 720p dvd movie or a 720p hd-dvd version of the same movie?

ah802
04-18-2006, 11:50 AM
Without a doubt the orginal 720p version should look better. Upscaling has to extrapolate (using some fancy foot work in chips) more information and spread it around a large display. Sort of like taking a 640x480 image and using a bi-cubic in photoshop to 1024x768.. vs the orginal that is already at that resolution.

That is if everything else is the same.. some upscalers are done well.

BobY
04-18-2006, 12:02 PM
Absolutely, it's not even theoretical. Although technically upscaling is really interpolation, not extrapolation (even if you are expanding the picture) as your are synthesizing pixels *between* the real pixels and not projecting beyond them.

A picture upscaled to a higher resolution will typically not be as sharp as a true higher resolution image because the interpolation smooths out the transitions between pixels, even those that shouldn't be smoothed (like hard edges--but some algorithms are smarter than others at that). Obviously the upscaled image won't have as much detail, as the detail wasn't captured to begin with and there is no way of synthesizing new details in the image (at least that would make any sense in the context of the rest of the image).

maicaw
04-18-2006, 12:55 PM
This is part of a post I put up a year ago on this other thread - an early technology used for high resolution compression utilizing fractal enhancement http://www.highdefforum.com/showthread.php?t=7471&highlight=fractal - I believe some modern codecs and scalers have been able to incorporate these computationally demanding algorithms to enhance video ] http://www.faqs.org/faqs/compression-faq/part2/ ...(Excerpt)Scalelessnes and Resolution Enhancement
When an image is captured by an acquisition device, such as a camera or scanner, it acquires a scale determined by the sampling resolution of that device.If software is used to zoom in on the image, beyond a certain point you don't see additional detail, just bigger pixels.
A fractal image is different. Because the affine transformations are spatially contractive, detail is created at finer and finer resolutions with each iteration. In the limit, self-similar detail is created at all levels of resolution, down the infinitesimal. Because there is no level that 'bottoms out' fractal images are considered to be scaleless. What this means in practice is that as you zoom in on a fractal image, it will still look 'as it should' without the staircase effect of pixel replication.The significance of this is cause of some misconception, so here is the right spot for a public service announcement.
/--- READER BEWARE ---\
Iterated Systems is fond of the following argument. Take a portrait that is,let us say, a grayscale image 250x250 pixels in size, 1 byte per pixel. You run it through their software and get a 2500 byte file (compression ratio = 25:1). Now zoom in on the person's hair at 4x magnification. What do you see? A texture that still looks like hair. Well then, it's as if you had an image 1000x1000 pixels in size. So your _effective_ compression ratio is 25x16=400.
But there is a catch. Detail has not been retained, but generated. With a little luck it will look as it should, but don't count on it. Zooming in on a person's face will not reveal the pores.Objectively, what fractal image compression offers is an advanced form of interpolation. This is a useful and attractive property. Useful to graphic artists, for example, or for printing on a high resolution device.[ed.--or display --]
That said, what is resolution enhancement? It is the process of compressing an image, expanding it to a higher resolution, saving it, then discarding the iterated function system. In other words, the compressed fractal image is the means to an end, not the end itself.

BobY
04-18-2006, 10:30 PM
I was thinking about fractals when I added the parenthetical "in the context of the image". I agree you could do hair, or blades of grass or other relativley randomized details, but fractals aren't going to logically or convincingly add rivets or portholes to your spaceship :D

maicaw
04-18-2006, 10:45 PM
but fractals aren't going to logically or convincingly add rivets or portholes to your spaceship :Dactually earlier in that same year ago thread-post#5 I made this observation - from a 1990 or so version - my version (of Iterated Systems) took hours on a 386 to compress a single image - the decompression was very quick - a few seconds on a 386.
In particular I took a picture of a WW2 bomber with just a few rivets noticeable in the original but the decompressed image showed them in greater detail and had a few more of them visible. -....

BobY
04-19-2006, 10:19 AM
How did it know where to put the rivets? Did it actually put them on the panel seams?

More importantly, how would it fair on a moving image? Would it always put the rivets where they were supposed to go on each frame as the object moved across the screen?

Fractals can be used effectively in generating detail in a CGI production where the computer knows in advance the parameters of the object it is mapping details onto, but I don't see it being able to add convincing details in real-time to moving images when it can't know anything about the objects on the screen.

Interesting idea, though, but it seems to me we'll have superior compression algortihms that will allow us to efficiently convey true Hi-resolution pictures with real details, before will have the processing power to add fractal details that make sense, in real time to a moving image.

maicaw
04-19-2006, 12:17 PM
How did it know where to put the rivets? Did it actually put them on the panel seams?.....Interesting idea, though, but it seems to me we'll have superior compression algortihms that will allow us to efficiently convey true Hi-resolution pictures with real details, before will have the processing power to add fractal details that make sense, in real time to a moving imageThe "new" rivets were in the proper locations - but the technique was crude - I scanned a drugstore print with a (1990s) flatbed scanner 600 dpi (3000x4200) - far better than the actual print resolution and then displayed the fractal enhanced image on the computer CRT (actual pixels) - impressive to me at the time - the rivets that were visible on the computer but not resolved on the drugstore print - may have actually been fully resolved on the negative -sort of a "latent watermark" kind of image in the print possibly - and the fractal compression "found" them - If I can find the original negative - I'll scan it to a 10 megapix TIFF with my digital-ICE negative film scanner and see what was actually on the negative - maybe even post a small detail image here - I recall that early versions of Encarta (Encarta 9x?) used hardware fractal compression on many images - despite the expense and difficulty because it provided realistic enhancement of the limited resolution available in computer photo-images -potentially disappointing to users who had seen the original versions in Funk & Wagnals print Encyclopedia -not far removed from the conversion of a bitmapped image to a vector (Corel or Postscript format) and the resulting ability to turn small pixelated detail into realistic enlarged geometric vector polygons Most real-world images have heterogenous mathematical properties; for instance a photograph in which mountains and clouds and trees might be represented by several classes of fractal representation. Barnsley's collage theorem proves that for a large class of real-world images, compact fractal representations must exist; it does not provide a general-purpose algorithm for the construction of such representationsactually fractal compression is attuned to "natural" patterns - wavelet compression works better for non-natural video images - my point was that for scenery and organic details - fractals work -
One frequent poster - here -pronounced (in his opinion) fractal enhancement a "fraud" :rolleyes: -

hdrichtv
04-22-2006, 04:53 PM
SD can never be converted to HD, I repeat, Never.

maicaw
04-23-2006, 09:34 PM
SD can never be converted to HD, I repeat, Never.PBS does it - everyday - and it says HDTV right there on the screen - and in the program listings - see this post by a PBS content producer http://www.highdefforum.com/showpost.php?p=69957&postcount=58 read this earlier and later posts to get the "context" http://www.highdefforum.com/showpost.php?p=66193&postcount=38 - here's the relevant quote Hi All,
I am the Director of Engineering at OPB, we produce History Detectives, it is produced in SD 16x9 and is upconverted at PBS

BobY
04-24-2006, 09:40 PM
We're speaking theoretically here when talking about fractals, but there is no reason at some time in the future we couldn't have DSP which added temporal resolution to SD video by intelligently synthesizing new frames in between real frames and which added visual resolution to SD video by intelligently adding fractal details. Absent a reference, the result would be indistinguishable from HD (and far better than upscaling, which can only reduce detail, not add it).

If that's a cheat or a fraud or not really HD, then recognize there is no such thing as HD. Everything we see that claims to be HD uses MPEG compression which is hardly a lossless algorithm, but it still can look good. Most HD we see is, in effect, computer-generated imagery, not an accurate reproduction of the original signal.

Audio purists would be aghast at a spectral comparison of a WAV recording and an MP3 made from that WAV, but people think MP3 sounds just fine given the benefits. HD video is no different.