At 8:48 PM -0500 12/13/04, Tom Barry wrote: >I think with proper deinterlacing maybe you could get slightly better >than .6. Yes, some people have suggest factors as high as 0.65 to 0.75. If you push the interlaced camera this hard, however, and you can do this in CCD cameras by adjusting the level of vertical detail, you will increase the perception of vertical twitter and other vertical aliasing artefacts. Thus it is a question of trade-offs - more detail versus more aliasing. > >But let's use your numbers. Say, instead of 1920x1080i we forget about >the silly "interlaced square pixels" and instead used something like >1440x1280i, still viewed in a 16:9 resolution. This has only 1,843,200 >pixels / frame or 921,600 pixels every 1/60 of a second. This is less >than 1080i and would be sending exactly the same number of visible >pixels per second as normal 720p. And at a slightly lower horizontal >scan rate. OK. Not sure what this accomplishes. Many HDTV products already subsample to 1440 or even 1280 samples per line for internal compression of 1920 x 1080@30i. You can play with resolution in any way that you want, but I would not call square samples silly, although some of their usefulness is lost when the format is interlaced. > >But the 1280 interlaced lines might, by your number, produce the >equivalent effective resolution of about a 1280 * .6 vertical resolution >of about 768 vertical lines. So, forgetting for the moment how badly >MPEG-2 might compress 4:2:0 interlaced chroma, we might get an effective >1440x768 resolution for the cost of 1280x720. Interesting. But in practice, I don;t think you would see any benefit. The filtering needed for using interlace would mitigate any potential benefit, and you would still need to deal with interlace artefacts, although they would be suppressed more than with 1080i. There would also be some implications for horizontal scanning rates in CRT displays, which would necessarily be higher than for 720P. The 48 kHz refresh for 720P is already a factor in the decision to run most CRT displays at 1080i (33.75 kHz). Pushing that rate higher is not going to help, but may not be important, since it looks like scanning CRTs are on their way out... in probably about the same timeframe as NTSC. > > Bottom line. There is no reasonable justification for the continued >> used of spatio/temporal undersampling for digital television. > >However I don't think interlace is the culprit here. It is the >obsession with square pixels that made 1080i so lopsided and thus >slightly short on effective vertical resolution. And extra horizontal >resolution that we rarely take advantage of. No, the benefits of square pixels are obvious in the other 1080 line formats...the progressive ones. It is interlace that is the culprit, since it requires more filtering to eliminate artefacts. I will agree that 1920 samples is overkill for 1080i, if for no other reason than the inability of CRT displays to deal with this level of detail. But as we move to fixed pixel displays, the difference between 1920 and 1440 samples per line should be quite obvious, if higher frequencies are actually contained the source. The real culprit is numerology and the CRT. The interlace trick works quite well on scanning displays, but it falls apart on fixed pixel displays, where de-interlacing is REQUIRED. Given the reality that the most affordable HDTV displays would be CRT based for many years, the Japanese ignored their own research and retained interlace in the NHK system. This allowed them to play the "1000 line" card, despite the fact that the system only delivered 1440 x 1035@30i, which has an effective vertical resolution of about 620 lines. We are still paying for this decision, made more than 20 years ago. You can carry on for days about the correct numbers for an interlaced HDTV format, but in the end, you will still have an interlaced NON HDTV format. There is no place for interlace artifacts in any HD programming. Furthermore there is NO NEED for interlace, as it has been proven that you can compress the "equivalent" progressive format more efficiently using digital compression techniques. What REALLY matters is delivering high quality samples that have not been trashed. As I have stated before for given bit rate, I can make better pictures (perceptually) using less spatial resolution, but more color resolution, while preserving the integrity of the samples. Take a good look at a 1024 x 576 RGB still image - this is adequate resolution to saturate the resolution capabilities of 95% of ALL TV displays now in existence, including virtually all of the HD capable displays. We are only now beginning to see mass produced fixed pixel displays that can do justice to 720P and beyond. But there is more to this story. If you are ONLY going to use this display to watch Television programming, you can play many games with, spatial/temporal resolution, anamorphic formats, et al. IF you want to also use this expensive display to view still images, run a graphical user interface, or view a web page, square pixels are EXTREMELY important, since they are the lingua franca of all that is IT driven today. Fixed pixel displays can deliver BOTH Nyquist filtered imagery and non-Nyquist filtered imagery; they can even overlay one over the other. Once we make the leap to progressive display of samples, making them square just makes good sense. Trying to keep the world of television DIFFERENT, as we move to an all digital infrastructure for the display of all kinds of imagery is what is REALLY "silly." >The Australians might have made a better deal of it with their 1440x1080i. Uhhhh...we got in on that deal too. It comes for free with MPEG-2. Just don't tell those guys at the ATSC that some of their beautiful 1920 x 1080 sources have been compressed to 1440 or lower, perhaps several times before they get delivered in all of their 1920 glory... ;-( The cable and DBS guys use 1440 or less all the time for 1080i source. Let's get over the numerology. The correct place to begin this discussion is with PROPER sampling and OVERSAMPLING techniques. Emission formats should be decoupled from acquisition and production; you get the greatest benefits in compression efficiency when you resample to a lower resolution emission format to remove entropy, and you get the best delivered images, when you run the compression algorithms in their "sweet spot" so that you do not trash the samples you are delivering. This is one case when LESS can REALLY be more. Regards Craig ---------------------------------------------------------------------- You can UNSUBSCRIBE from the OpenDTV list in two ways: - Using the UNSUBSCRIBE command in your user configuration settings at FreeLists.org - By sending a message to: opendtv-request@xxxxxxxxxxxxx with the word unsubscribe in the subject line.