How do you resize a video without losing picture quality?

[Wooooooo]

I have a clip here that I want to resize. The video clip I want to encode is currently at a 1280x720 resolution, 8000mbps, and picture quality is a sharp and clean image. I want to encode this to DVD specifications, resizing this video to 720x480 while maintaining the sharp clean imagery and without losing picture quality.

I encoded it once using TMPGEnc Plus and it produced a blurry picture seemingly lost some quality. I kept the bitrate the same at CBR 8000mbps since I’m not big on file size. My DC Component precision is kept at 10 bits. What’s the problem?

If you play a 1280x720 clip on Windows Media Player, then shrink the window size to 50%, the video itself shrinks and it seems cleaner and sharper. How do I get TMPGEnc Plus to output the current video looking like that?

[jagabo]

TMPGEnc uses a bilinear resize filter which gives blurry results compared to some other filters.

A good method of resizing is to use AVISynth's LanczosResize() filter. You can use FitCD to generate an AVISynth script, then use TEMPGEnc Plus to create the MPEG file from that script (ie, use AVISynth to resize, TMPGEnc just to encode).

[edDV]

You must realize that downsizing a 1280x720 progressive stream to 720x480 is going to reduce quality in the filtering process so you should be asking about least quality loss. By definition you will loose quality.

The issues go beyond downsize. If the image is 720p/59.94 fps and is film source as broadcast, it needs to be frame decimated in the proper 3:2 sequence to 23.976 as well as downsized to get to DVD 720x480/23.976p for DVD authoring.

If the image is 720p/59.94 fps and is live source as broadcast (e.g. sports, news, variety show, etc.), it needs to be converted to interlace as well as downsized to get to DVD 720x480/29.97i for DVD authoring.

[Alex_ander]

If your video is an interlaced MPEG2, it is important to separate fields, resize and weave them back. You have that control over video in Avisynth and VDubMod (or the more regularly updated VDubMPEG2). To use Avisynth you have to decode MPEG with DGIndex. Resize filter type is unlikely as important here (for reducing image size bilinear is good enough) but anyway there's a good choice (Lanczos3 in both cases, Lanczos4 and Spline36 in Avisynth 2.57).

[shootmeplz]

a clip at 8000mbps? are you serious? The movie i saw ordinarily is 800kbps more or less. Maybe it is me out of date.

[edDV]

a clip at 8000mbps? are you serious? The movie i saw ordinarily is 800kbps more or less. Maybe it is me out of date.

At 1280x720p, 8000Kb/s is about right for VC-1 on HD DVD. That is distribution quality for DVD. If from an on air ATSC MPeg2 source, 14-19 Mb/s is normal distribution quality.

The original was played at 100 or 144 Mb/s from tape.

[manono]

Resize filter type is unlikely as important here (for reducing image size bilinear is good enough) but anyway there's a good choice (Lanczos3 in both cases, Lanczos4 and Spline36 in Avisynth 2.57).

I might disagree with you that bilinear resizing is ever good enough. I would support jagabo that the TMPGEnc's bilinear resize is part of the problem. And if it's 1280x720, then it's probably (not guaranteed, but almost certainly) progressive in nature. By the way, for interlaced sources, just separating the fields and filtering before reinterlacing, the way VDub can do it, while much better than filtering the interlaced frames themselves, still blurs the video, and is inferior to smart bobbing, filtering, and reinterlacing, as is possible only in AviSynth.

I would also support edDV in stating that it needs to be either IVTC'd or made interlaced for DVD, depending on what the source is like, something else AviSynth is best at. If all Wooooooo did was to drop the video into TMPGEnc, then most likely that part of the conversion was screwed up as well.

If Wooooooo really serious about losing as little quality as possible when doing this kind of thing, I think he's better learn some AviSynth.

[Alex_ander]

...for interlaced sources, just separating the fields and filtering before reinterlacing, the way VDub can do it, while much better than filtering the interlaced frames themselves, still blurs the video, and is inferior to smart bobbing, filtering, and reinterlacing, as is possible only in AviSynth.

What is incorrect in unfold/resize/fold (VDub) for an interlaced source and what is different for resizing separate fields at smart bobbing? I can only understand it for filtering involving more than 1 field (like temporal noise filtering).

[jagabo]

Unfold/resize/fold causes problem because it ends up treating scanlines that are not next to each other as if they are. Consider scanlines 1 through 4:

1----
2----
3----
4----

unfolded become

1----2----
3----4----

Scanline 1 isn't really next to scanline 3 so it isn't really right to treat it as if it is. With blurry video the result usually isn't too bad but sharp horizontal edges or other small details can lead to nasty artifacts.

Shrinking a 720x576 frame to 720x480 with unfold/lanczos3resize/fold (as might be done for a PAL to NTSC conversion):



On top is the original, on the bottom is the resized result.

A smart BOB will attempt to leave the progressive parts of the frame intact and only deinterlace (by whatever means it uses) the interlaced areas. That reduces this type of artifacting.