gknot = best deinterlacing?

[chilledinsanity]

Hey I'm pretty tired, but I wanted to post this before I go sleep. It's been forever since I did an interlaced movie and I'm trying to do The Big Hit.

Is Gordian Knot's method of deinterlacing the best? And if so, which method, fast or field deinterlacing? If not, what's the better method? From what I read people seem not to like Virtualdub's method. Speed and size are no issue as I'm going to burn it onto DVD-R. Oh, if it makes any difference I'm encoding via Xvid.

I'm sure this has been asked before, but I'm too tired to read through all the search result. Sorry, thanks for the help

[JohnMK]

Field Deinterlace is just fine. I use it all the time to backup my pornos.

[jggimi]

I'll try to answer your question.

Gknot actually selects different AviSynth filters to deinterlace.

1) If you select Field Deinterlace, it will use Don Graft's filter of that name, from his DeComb suite of AviSynth filters. A very nice deinterlacer, though slower than Gknot's other choice.

2) Fast deinterlacing uses filters that cut the resolution in half, prior to resizing. This is very fast, but is usually not recommended unless you are dropping resolution appropriately. Ignoring cropping, this will cut a 720x576 or 720x480 DVD down to 720x288 or 720x240. If your final resolution is low enough (...x288 or smaller), then you might consider this option. (NOTE: The filter set changed at release 0.27, from VerticalReduceBy2 to SeparateFields/SelectEven. But the results are effectively the same.

3) There are other deinterlacers, such as GreedyHMA, that you can use by editing the .avs scripts.

For much more information, click on www.doom9.org/ivtc-tut.htm

JohnMK answered faster than I did, but I think you should still have a glimmer of the differences in order to make informed decisions.

[chilledinsanity]

thanks, that's very helpful

[piscator]

Originally posted by jggimi
2) Fast deinterlacing uses filters that cut the resolution in half, prior to resizing. This is very fast, but is usually not recommended unless you are dropping resolution appropriately. Ignoring cropping, this will cut a 720x576 or 720x480 DVD down to 720x288 or 720x240. If your final resolution is low enough (...x288 or smaller), then you might consider this option. (NOTE: The filter set changed at release 0.27, from VerticalReduceBy2 to SeparateFields/SelectEven. But the results are effectively the same.
JohnMK answered faster than I did, but I think you should still have a glimmer of the differences in order to make informed decisions.

I was under the impression that resolution does not affect (or reduce) quality for this option. I worked a few years ago on such interlaced algorithms for my job and what I recall from interlaced formats is that a single frame consists of an odd field and an even field. The odd field contains the odd lines and the even field contains the even lines. Thus, this filter should create a new non-interlaced frame from both fields.

It is true that the PAL definition contains 288 lines (thus 144 lines per field). But I suppose if the input resolution contains 576 lines, both fields would contain 576/2=288 lines. Thus not reducing quality if your target resolution is bigger than XXX x 288.

btw, some television sets can also show the interlaced effect. If you have a 'freeze screen' option, you can notice that only half of the lines are displayed.

greetz,
Piscator

[jggimi]

Thus, this filter should create a new non-interlaced frame from both fields.I may be wrong, but I don't think so for this particular filter set. Prior to 0.27, the filter was VerticalReduceBy2 (http://www.avisynth.org/index.php?page=ReduceBy2)....VerticalReduceBy2 reduces the vertical size by half...At 0.27, it was replaced by two filters, SeparateFields (http://www.avisynth.org/index.php?page=SeparateFields) and SelectEven (http://www.avisynth.org/index.php?page=Select).SeparateFields takes a frame-based clip and splits each frame into its component fields, producing a new clip with twice the frame rate and twice the frame count....SelectEven makes an output video stream using only the even-numbered frames from the input. This is why "fast" deinterlacing is so fast, and also, why it should be used with care.

[EDIT: You don't have to believe me, just edit any "fast deinterlace" AviSynth script, and delete or comment out the resizing filter, then open it in Vdub or any AVI player.]

[piscator]

Thanks for the links jggimi.
Looking at the description of SeparateFields, I think they precisely do what they should do for deinterlacing.

Note that the vertical resolution is not cut in half, but the target frame rate and frame count is cut in half compared to the source field rate and field count. A single target frame is constructed from two source fields: one field contains the odd lines and one field contains the even lines. The target frame will contain all the lines.

Looking at the description of VerticalReduceBy2, I think you are right: the vertical resolution is cut in half. Basically, the unused lines in every source fields are discarded. Note here that not 1 frame is "weaven" from two source fields. This is a very crude deinterlace method and requires the target vertical resolution to be cut in half too (if you don't the quality is not better anyway).

Fortunately I only used GKnot 0.27, so I didn't mess up my compressed movies :-)

greetz,
Piscator

[jggimi]

I think you may have misunderstood me.

First, you are correct, in that SeparateFields does separate a frame into it's component fields. It doubles the frame count, but each new frame has 1/2 the resolution of the original content.

Then, the SelectEven command selects every other frame -- which will be one of the two fields. Unfortunately, that field has 1/2 the vertical resolution of the original frame.

This has the same effect on image size as VerticalReduceBy2.

Allow me to prove it. Here are two images -- which I've run through the process, unresized, so you can see what I mean.

The first is an original frame (560x320) that I saved in an .avi file.

The second is the same frame, run throughAviSource("test.avi")
SeparateFields()The resulting .avs has 2 frames, each 560x160.

You'll find them in the attached .zip file.

NOTE: If you add SelectEven() to the script, you'll end up with 1 frame, rather than 2, at 560x160.

[piscator]

Thanks for the example jggimi. It proves your point undisputedly. Apparently, I didn't read the discription of SeparateFields careful enough.

greetz,
Piscator