how to fix hard-encode pillarboxing?

[spiritgumm]

planning to re-encode (to dvd) this dvd recording from TV because of the telecine pattern (which is not part of my question), and wondered if I should also crop borders and resize? or better to leave that as is?

[jagabo]

Encode for what? DVD? For DVD I'd crop to 540x480, resize to 720x480, and encode as 4:3 DAR. Actually, I'd probably crop away a little more then resize to 704x480 to get rid of the bright column of pixels at the right border.

[LMotlow]

Crop(102,0,-98,0)
Spline36Resize(704,480)

Re-encode the results with 4:3 DAR.
Are you aware that your deinterlaced video has duplicate frames?

[spiritgumm]

I previously tried cropping/resizing, but previewing on my PC, the pictured looked too stretched out, so I started cropping less and less, then I thought I should just ask in case I was suppose to resize to a much larger width (like 853) and then crop. That was the suggestion in another thread.

As mentioned, the frame pattern is not the question (and the clip is the original video, not deinterlaced), but for the sake of clarity, it's got judder from 2224 telecine which I want to change to 2:3 pulldown.

[LMotlow]

I cropped and resized as well, and made two versions: "A" is 704x480 telecined (3:2 pulldown) for DVD. "B" is 640x480 progressive 23.976. Neither looks stretched to me -- they both play as 4:3.

Hate to say it, but your original sample has motion problems and serious banding issues that need cleanup. Look at the background wall that Sterling Hayden walks along. Banding and blocking make the wall textures change shape. Whoever removed all the original film grain caused most of that noise. Has a lot of spots, too.

TDecimate()                 #<<<- (remove duplicate frames)
Crop(102,2,-98,0).AddBorders(0,0,0,2)   #<<<- (crop and center vertically)
RemoveSpotsMC()
GradFun2DBmod(thr=2.2,mask=false)
Spline36Resize(704,480)           #<<<- Or 640x480 for the mp4
LimitedSharpenFaster()
grainfactory3(g1str=2, g2str=2, g3str=2)
AddGrainC(1.5,1.5)

[spiritgumm]

you probably needed a different clip to see people and circular objects looked abit squat at 720 on a PC monitor. I guess temporarily resizing to 640 for PC would be more accurate to get an idea if the video looked normal.
I dont know if the original ripper did anything besides record from TV, and whether they used a Tivo or dvd recorder.

[LMotlow]

you probably needed a different clip to see people and circular objects looked abit squat at 720 on a PC monitor. I guess temporarily resizing to 640 for PC would be more accurate to get an idea if the video looked normal.

A 720x480 video when not displayed at the correct aspect ratio will not display correctly. 4:3 will look stretched and squat, 16:9 will look tall and thin. Always.

If you're looking at uncompressed/decoded AVI, there is no aspect ratio flag. It will display as 720x480 or 720x576, or at its physical frame size, whichever applies. You can always adjust the display in Virtualdub and most editors to display the correct aspect ratio for the video.

[jagabo]

The only ambiguities here are whether the source uses ITU or MPEG aspect ratios, and whether you want to use ITU or MPEG aspect ratios in your final output. With ITU scaling the DAR is contained in a 704x480 portion of the frame. With MPEG scaling theDAR is contained in the full frame. The DVD spec refers you to the MPEG spec regarding aspect ratios. The MPEG spec is clear -- the full frame contains the DAR.

When standard definition analog video tape is captured it is done to the ITU spec (all professional digitizers follow the ITU spec) -- the 4:3 image is captured in a 704x480 frame and an extra 8 pixels is captured at the left and right edges in case the video or cap is off center. Rather than adjusting the ITU cap to match the MPEG 2 spec the 720x480 frame is usually sent directly to the MPEG 2 encoder and encoded as 4:3 DAR. Ie, nobody cares about the ~2 percent difference between the two specs.

If you want to follow the MPEG 2 spec: crop the 720x480 source down to 540x480, resize up to 720x480 and encode as MPEG 2 with 3:4 DAR. If you want to follow the ITU spec: crop the frame down to 528x480, resize to 704x480, and encode as MPEG 2 with 4:3 DAR.

[LMotlow]

Yep. Makes sense to me. As you mention, from the sample it's difficult to tell where the original borders were -- i.e, did the original 4:3 image exist as 704x480 with 8-each pixels at the sides, or the equivalent of 16 border pixels somewhere? I've seen 4 pixels on one side, 12 on the other, etc. So my script took the lazy way out and assumed zero side border pixels, and cut 720x480 down to 520x480. Resized to 704x480, that makes for a slightly wide image at 4:3 DAR. In my own VHS caps (usually at 640x480), I go thru similar arithmetic with those borders (but not exactly the same) and leave some border pixels to get 720x480 with 4:3 DAR. I'll have to play with my numbers on a recent cap to make sure I'm getting the precise image on playback. But I see what you mean.

Maybe some cropping going on with that movie sample? I vaguely recall that film from a long way back, seems it might have been 1.6:1 ? ? But I couldn't locate reference to the movie anywhere, so I'm not certain. Wonder where the top border pixels came from? Anyway, most Hollywood originals shot at "4:3" were really shot on 35mm film at 1.37:1, which is slightly wider than 4:3. A lot of what I see on old TV movies that was shot that way looks either cropped or a bit squished. So regardless how I capture 4:3, I do some manipulation to get a 1.37:1 image on playback from 720x480. Sometimes. Sometimes it ain't worth it.

Saving the previous post in my jagabo folder for reference. Thanks.

[jagabo]

Keep in mind that back in the days of SD TV it was typical for a CRT TV to have a 5 to 10 percent AR error. Hardly cared about that.