It all comes down to the fact that DVD is more of an analog format than a digital one. I feel like people get “CD ripping brain” which causes them to think that the most desirable thing is making the most-accurate copy of what's on the disc. For CD that's true because PCM is PCM, but for DVD the thing we really want is the program material, which is three layers deep on a DVD: inside an NTSC video signal, which is digitized following the Rec.601 standard, which is then shoved into an MPEG2 transport.
Four major things that can be done to DVD to make them look great on modern displays:
- Deinterlacing is the hardest to get right. Progressive-scan 24-frames-per-second DVDs exist but are mostly confined to movies where there will be a better BD release anyway. Interlaced DVDs where the program material is intended to be seen in 24FPS get “inverse telecine” (IVTC) instead of straight deinterlaced, but again I don't do a lot of those for the same reason. Almost any NTSC DVD that I care to encode is thus going to be 60000/1001 fields per second, which needs to be turned into 60000/1001 frames per second to avoid throwing away half of the available motion detail. If you do nothing at encode-time and produce an interlaced output, then the display or player software will end up doing it and will do a bad job. HandBrake's deinterlacing options just don't look good in my experience. I like QTGMC for this because it predicts the motion of the infill fields instead of just copying the previous field verbatim. It's very noticeable any time there's a lot of horizontal movement in the program material.
- Resolution and ratio. Most people hear “anamorphic” DVD and think of 16:9 crammed into a 4:3 image, but the truth is that all NTSC DVDs are anamorphic. They're 720x480 which if you calculate it is actually a 3:2 aspect ratio. Very clever because it ends up being about the same amount of scaling for 4:3 or for 16:9 material. They rely on PAR/DAR flags to tell the player or display how to scale it, but modern displays have terrible terrible scalers because it's purely a box-checking thing for them and not a feature they spend money or effort on. When I encode a DVD I stretch it myself at encode-time to 720x540 or 960x540. There's obviously some artifacting inherent in that vertical stretch, but it avoids throwing horizontal resolution detail away by scaling 4:3 programs down to 640x480 like most encoders do. Then the 540 pixel-doubles cleanly into 1080, 2160, etc.
- SD colorspace (Rec.601 again) is a similar issue where modern displays are just fucking terrible at it because there's no economic reason for them not to be. The chroma is already subsampled, so greens especially end up looking washed out and terrible. When I encode a DVD I convert them into HD colorspace which doesn't restore subsampled chroma but at least avoids letting the display make it worse.
- Cropping. The program-area resolution is actually 702 or 704x480 for anything transferred from tape (look up SONY D-1). If you have any "DVDrips" sitting around of an '80s or '90s TV show, does it have 8 pixels of black pillarbars on the left and right? If so then the person who encoded it didn't know what they were doing. It subtly throws off the aspect ratio for the entire program, especially noticeable in animation where they tended to use exact-circle tools. Look at the characters' eyes in The Simpsons for a great example. I crop those off before my one-time scaling so the program ratio comes out perfect.
This all applies similarly to PAL DVDs except I'm usually shrinking them down to 540px because the loss of some vertical resolution is still better than trying to get a modern display to scale 576px to panel-native res, and deinterlacing PAL is a straight 50-fields-to-50-frames without the wacky 1001 division notation that is a legacy of the backwards-compatible way that color was introduced to NTSC.
alt Hacker News
It all comes down to the fact that DVD is more of an analog format than a digital one. I feel like people get “CD ripping brain” which causes them to think that the most desirable thing is making the most-accurate copy of what's on the disc. For CD that's true because PCM is PCM, but for DVD the thing we really want is the program material, which is three layers deep on a DVD: inside an NTSC video signal, which is digitized following the Rec.601 standard, which is then shoved into an MPEG2 transport.
Four major things that can be done to DVD to make them look great on modern displays:
- Deinterlacing is the hardest to get right. Progressive-scan 24-frames-per-second DVDs exist but are mostly confined to movies where there will be a better BD release anyway. Interlaced DVDs where the program material is intended to be seen in 24FPS get “inverse telecine” (IVTC) instead of straight deinterlaced, but again I don't do a lot of those for the same reason. Almost any NTSC DVD that I care to encode is thus going to be 60000/1001 fields per second, which needs to be turned into 60000/1001 frames per second to avoid throwing away half of the available motion detail. If you do nothing at encode-time and produce an interlaced output, then the display or player software will end up doing it and will do a bad job. HandBrake's deinterlacing options just don't look good in my experience. I like QTGMC for this because it predicts the motion of the infill fields instead of just copying the previous field verbatim. It's very noticeable any time there's a lot of horizontal movement in the program material.
- Resolution and ratio. Most people hear “anamorphic” DVD and think of 16:9 crammed into a 4:3 image, but the truth is that all NTSC DVDs are anamorphic. They're 720x480 which if you calculate it is actually a 3:2 aspect ratio. Very clever because it ends up being about the same amount of scaling for 4:3 or for 16:9 material. They rely on PAR/DAR flags to tell the player or display how to scale it, but modern displays have terrible terrible scalers because it's purely a box-checking thing for them and not a feature they spend money or effort on. When I encode a DVD I stretch it myself at encode-time to 720x540 or 960x540. There's obviously some artifacting inherent in that vertical stretch, but it avoids throwing horizontal resolution detail away by scaling 4:3 programs down to 640x480 like most encoders do. Then the 540 pixel-doubles cleanly into 1080, 2160, etc.
- SD colorspace (Rec.601 again) is a similar issue where modern displays are just fucking terrible at it because there's no economic reason for them not to be. The chroma is already subsampled, so greens especially end up looking washed out and terrible. When I encode a DVD I convert them into HD colorspace which doesn't restore subsampled chroma but at least avoids letting the display make it worse.
- Cropping. The program-area resolution is actually 702 or 704x480 for anything transferred from tape (look up SONY D-1). If you have any "DVDrips" sitting around of an '80s or '90s TV show, does it have 8 pixels of black pillarbars on the left and right? If so then the person who encoded it didn't know what they were doing. It subtly throws off the aspect ratio for the entire program, especially noticeable in animation where they tended to use exact-circle tools. Look at the characters' eyes in The Simpsons for a great example. I crop those off before my one-time scaling so the program ratio comes out perfect.
This all applies similarly to PAL DVDs except I'm usually shrinking them down to 540px because the loss of some vertical resolution is still better than trying to get a modern display to scale 576px to panel-native res, and deinterlacing PAL is a straight 50-fields-to-50-frames without the wacky 1001 division notation that is a legacy of the backwards-compatible way that color was introduced to NTSC.