Recognizing Upscaled HDTV

[minolta]

When non-HD material is broadcast as HDTV, it appears to be "bicubic" resized to HD resolution. This can make it difficult to determine the original resolution. Consider the following caps taken from Leno, Raymond, and Dave:

http://www.vrac.iastate.edu/~chadspen/leno.jpg
Leno: 1920x540 field on left; 740x240 field on right.

http://www.vrac.iastate.edu/~chadspen/raymond.jpg
Raymond: 1920x1080 decombed frame on left; 720x240 frame on right.

http://www.vrac.iastate.edu/~chadspen/dave.jpg
Dave: 1920x540 field on left; 740x240 field on right.

The left image is cropped from the original HD, while the right image is first Lanczos downsized to 720x480 and then Lanczos upsized to the original 1920x1080. I know this is not very scientific, but its an easy way to compare.

All caps were taken with things in focus and little movement. The Leno clip is noticeably blurred by this operation. Since the show is indeed shot with 1080i cameras this is not surprising. The Raymond clip is also widescreen HD, but does not appear to be affected (not much, anyway). This did not surprise me, since this show has always appeared blurry or smoothed. Perhaps its the camera filters they use, or could it be upscaled from DVD quality? Last is Dave, which is almost certainly upscaled from NTSC (although, overly strong bicubic w/ ugly color distortion).

Worse yet, movies broadcast in HD often look like upscaled DVD as well. Is there any way to "scientifically" determine the effective resolution? I've tried running tests using avisynth's Compare(), but the results don't look too promising. Is there any such measure?

Thanks,
Minolta

[trbarry]

See my AVS Let's measure detail (http://www.avsforum.com/avs-vb/showthread.php?s=&threadid=242229) thread from last year. It's the closest I've been able to come to it.

It basically resizes down to DVD resolution, then back up, and compares to see how much was lost.

- Tom

[minolta]

Thanks Tom, great thread. I knew this issue had be bugging someone other than myself...

I tried in vain yesterday for 3 hours trying to come up with a quick method for estimating resolution, but of course failed. I did the resize-down/resize-up at various resolutions and plotted the results. I fully expected the PNSR or MAD to plateau around the original resolution, but I guess it's trickier than that. There must be some mathematical way to estimate the effective resolution. It is certainly an interesting problem, at least.

So far, I guess the best way is to resize/upsize and just eyeball it. If it looks and smells like DVD quality, then it probably is (or it's so filtered that it just as well be).