> 1) The optical anti-aliasing filter should eliminate any aliasing from > optical anamorphic. Only to the extent it alerady does so in isomorphic 4:3 production. Some cameras are better than other at this. > 2) Nevertheless, FINE detail may be optically compressed to SUPER FINE > detail which will be filtered out by the anti-aliasing filter. No more so than zooming wider to the same horizontal angle of view. > "non-oversampled" CCD cameras may well lose detail toward the edges that > won't be restored when the image is stretched horizontally. In short, in > a prosumer video camera, optical anamorphic compression is lossy. It won't be towards the edges unless the lens is poorly designed, it'll be a 25% reduction in horizontal resolution over the entire image. At the same horizontal angle of view, it'll be exactly as lossy as "fake" 16:9 since the same number of pixels horizonatlly are being used in either case (this assumes no MTF reduction from the anamorphic adapter itself). In both cases resoltion is reduced 25% (remember resolution is normalized to active picture height). > 3) The optical anamorphic lens may itself contribute to overall image > softness. Yes. This is the biggest problem with this approach. > 4) Unless the camera has a CRT viewfinder that can be adjusted for > vertical height, composition is going to problematic with optical > anamorphic. Oddly enough, one adjusts to it. But yes. > THEREFORE, optical anamorphic is not a great solution. And it costs > money. Er, ah, it's got tradeoffs, just like anything else. It's not great, but then, neither is shoot-and-protect 16:9 in the 4:3 frame nor "fake" 16:9. And even shooting true 16:9 (DSR-500WS, AJ-D610WA, etc.) is annoying: it's still SDTV. Now, what you really need is an HDW-700A -- or an F900 for proscan. Hey, it's only money! :-) > 5) Electronic 16:9 that uses LINE REPLICATION (every 4th line is > duplicated) will cause a pre-compression loss of 25% vertical > resolution. I can't think of a single system currently available (some cheap PC scan converters aside) that operate through line replication or decimation. Every "fake" 16:9 camcorder I've seen does multitap upsampling interpolation > 6) Contrary to Perry, I believe that line INTERPOLATION (using the line > above and below the "missing" line) can create lines with information > that is very similar to what the "missing" lines would have been if > captured. (That's why jaggies are reduced.) While INTERPOLATION can't > double vertical resolution, it could be expected to create a 25% > increase in pre-compression resolution. This could make 16:9 mode no > worse than letterboxing a 4:3 image. The jaggies are reduced because high-frequency information has been lost. You cannot make missing detail out of thin air, but you can upsample smoothly so as to prevent jaggies and render a pleasing image. Apparent increases in detail can be attributed to lower losses going through the DV codec, since the upsampled image stresses the codec less. > 7) Prior to compression, the 360 lines are not simply stretched over > 480-lines as is claimed. Whether INTERPOLATION or LINE REPLICATION is > used to expand the image, there are 480-lines of information to be > compressed. Well, yes, but the interpolation IS stretching the 360 lines to 480 lines. > Thus I can't really see any difference between compressing > these 480-lines using 60 blocks or 360-lines with 45 blocks. The > compression load is almost the same. The total useful data, as it were, is the same, but on the block level, the stretched (fake) image reduces the detail in each 8x8 pixel block by 25%, reducing compression difficulty. Better than shoot-and-protect 4:3 if all you want is the 16:9 center panel, but a no-go if you also need to release in 4:3. But shooting anamorphic uses *all* 480 scanlines to capture the image and is neither better nor worse horizontally for the same final angle of view (all else being equal, which as discussed is not always the case). > 8) No one is sure exactly how vertical resolution is compressed and > restored. The images I see are consistent with at least a simple linear interpolation filter. These are trivial to implement either in software or hardware since there's no need for a general purpose solution, only four sets of fixed coefficients (one set per line in the repeating 3->4 line upsampling cycle). But no, I don't know how they're doing it. > 9) In any case, we agree that electronic anamorphic couldn't BY ITSELF > cause a huge INCREASE in vertical resolution as shown in the screen > shots. Yes. >(I would add, that if true line interpolation was used, I could > see how a STATIC res. chart could show some increase!) Only when accounting for improved codec efficiency due to lower complexity in each 8x8 block. > 10) If row-summation were turned off when in 16:9 mode, resolution could > increase. But there would be noticeable side-effects! Grab a DSR-300 or DSR-500 and turn on "enhanced vertical sharpness." Or pick up a TRV900, VX2000, etc. and switch it to proscan (lock it down on a static shot to avoid temporal complications) You'll see it! Sensitivity drops one stop, and the static image is both sharper vertically and more prone to jaggies. And on an interlaced monitor, especially with sharpness cranked, it can be VERY painful to watch: flicker, flicker! > The question is how much difference is there between a > 500WS in 4:3 mode and a 300 shooting in its natural (4:3) mode. If there > is little difference, then buying a switchable camera is a good deal. So > what's the resolution of the 300? DSR-300: 800 TVL/ph DSR-500: 750 TVL/ph In practice, the pix are very close, but with sharpness up I can detect just a trace of horizontal aliasing on the 500 on certain details whereas I can't on the 300 on the same scenes. But the difference is very subtle and difficult to see in 99% of subject material. > Here is an alternative 16:9 solution. Mark ONE line on your 4:3 > viewfinder TOWARD THE BOTTOM -- 75% from the TOP. Now shoot in regular > 4:3 mode composing with the left/right/top edges -- leaving "fluff" > below this line. Or crop some off the top and bottom. I've seen shoot-and-protect done well both ways -- and screwed up both ways! > I've done this in real-time with the DigiSuite. I don't know if the > C-cube based products can do this -- meaning I know they don't Well, the DTV uses the C-Cubes, grin... but I get your drift. For the non-Matrox C-Cube set (DV500 etc.), simply build a cropping matte graphic and put it in the other channel as a super. Works fine. I've also done it in real time with a double-edged wipe in the WJ-MX50 (any other vision mixer will do as well) -- which also, with its compression option, lets me make letterboxed 4:3 versions of vertically stretched material -- great for quick-n-dirty demo tapes or 4:3 workprints from widescreen DSR-500 footage or just for getting a feel for widescreen material on 4:3 monitors. > I make no claims that this solves the vertical resolution issues, BUT it > does mean you can get 16:9, 14:9, and 4:3 from the same source material. > And it will work with ANY viewfinder. Moreover, it solves the "common > sides" issue when editing for different aspect ratio releases. On this, finally, we agree completely! :-) Cheers, Adam Wilt
> However after numerous posts on this lists, the consensus seems to be > the "fake" 16:9 is not that bad after all, and certainly far superior > to cropping in post. Well, it gives better quality than cropping in post, but you lose the option of a full-frame 4:3 release. TANSTAAFL. Cheers, Adam Wilt