Does halving the height of a video deinterlace it?

[pon pon]

I have an interlaced 1440x1080 video and I want to make a smaller deinterlaced AVI version of it. Am I corect in assuming that if I halve the vertical resolution to 540 I am effectively deinterlacing the video at the same time? This is based on my (recent) understanding of interlacing, that each frame only has half the horizontal lines of the image and combines with an adjacent frame.

If this is the case, then it seems to me to be a great way to deinterlace, avoiding lossy deinterlacing algorithms. Even as I write this, this notion of mine seems too good to be true. A brief and simple explanation as to why this is not the case would be appreciated, if the answer to my question is no.

[Guest]

Yes, that is a deinterlacing method. It's disadvantages are:

1. The video is now half the original height. If that is what you wanted, then fine. If not, well that's serious.

2. The aliasing will be poorer compared to other possible methods. Only you can decide if it is too poor.

[pon pon]

Thanks for the quick reply. I do not understand your point 2 though. I thought doing this would preserve the original image data because halving would only remove the "empty/blank" alternate lines. Interlaced frames in this example only have 540 lines of horizontal data, don't they?

It seems I really do not understand interlacing after all.

[Guest]

Suppose you have a static scene with a diagonal line. I say static so that there is no movement between the fields. Then if you weave the two fields to make a full frame then you have a nice clean line. If you take just one of the fields you will have a "lumpy" aliased line. There are ways to process things such that the line is not as lumpy.

[CWR03]

Interlacing isn't "blank" lines interlaced with the video, it's video drawn on the screen in the opposite direction. Half-height deinterlacing is going to take away every other line of your image, squeeze them together for "storage" and re-stretch them to the proper display aspect ratio on playback. When you do this, you literally throw away half of the image detail.

[Guest]

He means that when a field is displayed on an interlaced device, only half the lines are displayed for the full picture size. So he's not really wrong about that. On the other hand, when you want to make full frames for progressive presentation, then, as you say, the fields can be directly weaved, which would produce combed frames, or you can more intelligently combine them trying to eliminate combing while retaining as much of the information from both fields as possible.

That's clarification for the OP as I know CWR03 fully understands all this. :)

[pon pon]

Losing half the horizontal image data is not a problem. In fact, it is desirable because the video is way too big for my purpose.

Reading the most recent comments and thinking about this some more, what I was suggesting was generating a new frame from only half the fields, whereas an alternative is using a "more intelligent combination" of both fields, which I take to mean a sophisicated deinterlacing algorithm.

The complication as I understand it with conventional deinterlacing is the combination do not represent the same point in time. There is a time shift of 1 frame with half the data which the algorithm attempts to correct. So is it better to produce a new frame from half the data that is 100% time-correct, or use twice as much data that is "nearly" time-correct? Even if the end resolution exactly matches the first method I now suspect it is the latter, which must be why so much effort has gone into various deinterlacing algorithms.

[Guest]

Don't forget that a picture may have static areas and changing areas, so your choice is not necessarily a Hobson's choice. You can weave in static areas and get the full resolution, and interpolate in the changing areas. So that is one "more intelligent" way.

Even if you are going to half height, just throwing away every other line is equivalent to point resizing, which is the worst way to resize.

You should consider using a good ELA or NNEDI deinterlace followed by a decent resizing kernel.

[pon pon]

I will look into your suggestion. Thanks.

It is clear that you are no fan at all of deinterlacing by halving the height. I was attracted to this idea because I figured there was no actual resizing and no pixel distortion taking place in the deinterlacing process. (Similar in my way of thinking to anamorphic encoding.) But it begs the questions: "Exactly how accurate are those undistorted pixels in representing the original image in the frame? How well does one set of fields do that?"

I have spent way too much time in the last 5 hours thinking about deinterlacing!

[Guest]

But it begs the questions: "Exactly how accurate are those undistorted pixels in representing the original image in the frame? How well does one set of fields do that?" I didn't beg it. I told you that you will have more aliasing with the point resizing. If you think a lumpy line represents the original clean line better than one made from a better process and is therefore more "accurate" then by all means go for it.

[pon pon]

Sorry, I wasn't intending to suggest that you begged those questions. It would have been clearer if I had written "But my original idea begged the questions:..."

Well, it is 5am where I am now and I desperately need some sleep. That is my excuse.