Creating High Res from standard input

[wgscott]

I just read this:  Arse Technica: We can now convert every film and TV show from the last 80 years into HDR

 

It seems to do the right thing.  Presumably audio is an easier target.  Could an analogous approach be used to expand dynamic range in a non-artificial way?

[Cebolla]

Is this bottom talk or just a Brit Freudian slip / genuine typo?

[wgscott]

OK, ok, so I have made the joke a few times before.

[esldude]

You haven't heard of HMDQRA?

[mansr]

5 hours ago, wgscott said:

I just read this:  Arse Technica: We can now convert every film and TV show from the last 80 years into HDR

 

It seems to do the right thing.  Presumably audio is an easier target.  Could an analogous approach be used to expand dynamic range in a non-artificial way?

I'd expect audio to be harder. It is usually much easier to play tricks with visuals and get satisfactory results than it is with audio. For whatever reason, our auditory perception is much more sensitive to distortion. Modern video codecs achieve compression ratios of 200:1 while still looking great. The audio equivalent would be compressing CD quality to 8 kbps and expecting it to sound nearly as good. Although the article isn't about compression, this still shows how much more easily our eyes are tricked than our ears.

[gmgraves]

4 hours ago, wgscott said:

I just read this:  Arse Technica: We can now convert every film and TV show from the last 80 years into HDR

 

It seems to do the right thing.  Presumably audio is an easier target.  Could an analogous approach be used to expand dynamic range in a non-artificial way?

 Sure, things can be enhanced, but "converted to HDR" is stretching reality more than a little bit. Yes, the examples are impressive, but I know that still images can be improved like that using Photoshop or GIMP. The article is right, while software can do a magnificent job removing scratches and repairing damaged frames in film, increasing contrast, and restoring color to film and videotaped images, it can't add more information to an image than was capture by the media originally, a 35mm cine frame for instance has roughly 75 lines of resolution/mm (best case), Since a "line" of resolution is defined in photographic terms as one light-to-dark cycle, you have to allow 2 pixels for that so we have to double that figure that to roughly 150 lines (for counting purposes).  That works out to 3300 pixels X 2400 pixels for an 22mm X 16mm standard Academy aperture cine film frame. One can double-up on pixels by "up-converting", but while that might increase "apparent" resolution, it won't add any more detail to the image than was already there. Video images, whether on video tape or kinescope were, for the bulk of the "video age" was 486 lines  of vertical (for NTSC's 525 line system) X 720 pixels of horizontal resolution. PAL had more vertical lines but the same 720 pixels of horizontal resolution as was dictated by a 6 MHz video bandwidth. Again, line doubling and converting to progressive scanning rather than interlaced can increase apparent resolution, but no more actual video content is added. Also, realize that these figures don't take into account the limitations on image quality inflicted by lens resolution, contrast, and such lens restrictions as spherical aberration, pincushioning, barrel distortion, chromatic aberrations, etc. 

 

The same is true with audio. You can't get any more off of a phonograph record or an analog audio tape or a digital recording than is there. You can do some things to enhance a recording somewhat, but ultimately, you are going to encounter trade-offs that at some point become unacceptable. For instance, boosting high frequencies to obtain more output at the point where the tape's frequency response starts to fall-off, merely increases noise and distortion. Analog tape recordings have very little on them above 15 KHz at best, and often have less due to self erasure over the years. Recording studios could not and do not maintain their analog tape recorders above 15 KHz because no alignment tapes were ever made that went higher than that. The reason is very practical. Tape-to-head contact deteriorates with rising frequency. One can hardly get a 15 KHz tone steady enough to do head alignment, much less to get an accurate EQ measurement. Of course this situation can be ameliorated to some extent by increasing the working  linear tape speed.  The bottom line is, of course, don't look for much output of even a pro recorder, running at 30 ips (76.2mm/inch) above 15 KHz. It is doubtful that a copy of Getz/Gilberto (which was mastered by Verve on 35mm magnetic film) downloaded from HDTRacks takes much advantage of that 48 KHz audio bandwidth afforded by the 96 KHz sampling rate, but at least you know that you have a good chance of that copy containing every last iota of information available on the master tape. 

[patagent]

22 minutes ago, mansr said:

I'd expect audio to be harder. It is usually much easier to play tricks with visuals and get satisfactory results than it is with audio. For whatever reason, our auditory perception is much more sensitive to distortion. Modern video codecs achieve compression ratios of 200:1 while still looking great. The audio equivalent would be compressing CD quality to 8 kbps and expecting it to sound nearly as good. Although the article isn't about compression, this still shows how much more easily our eyes are tricked than our ears.

Images tend to be less complex (more "sparse") than music.  You can regularly beat Nyquist theorem in imaging using compressive sensing principles.  Reconstruction of music based on compressive sensing only works for the simplest signals:

 

http://sunbeam.ece.wisc.edu/csaudio/