100fps: How to deinterlace Video the right way. Examples, Tricks, Comparisons

OK, here are some file sizes for you, to show you what to expect with different kinds of deinterlacing and DivX compression methods.

 

I compressed a 42 minute movie with a lot of fast moving scenes.
I used DivX Professional 5 with "Constant Quality" of 93% (but that's up to you),
keyframe interval "max every 50 frames", "GMC" and "Bi-directional Encoding".
I encoded with "No sound", so that the figures below are purely video.

Click here to see how the movie compresses with HuffYUV and other (nearly) lossless compression methods.
Click here to compare encoding of fine lines that tend to quiver.

 

fps

Method

%

Important Details

(1) 50 fps
720x288
Field Bob
(Method 4a)

100.0%

Virtualdub's "Sharpen" filter set to 10 to compensate blurring of the "Field bob" filter.
Compared to an uncompressed movie this is about 1:106. It's fascinating to see what DivX is capable to do without nearly any visible loss.

This could be the deal for you. However for best results use method 15.

(2) 50 fps
720x288
Field Bob
(Method 4a)

97.1%

The same as (1) but keyframe interval is set to "500".
3% saved without sacrificing quality is not bad. The drawback: Seeking (fast forwarding/ rewinding) takes longer.
(3) 25 fps
720x576
Blend

 86.7%

 
Don't do it. Result looks bad. You lose details and motions are unnatural. This method is called "(best)" in Virtualdub while in my opinion it's not, but it's a matter of taste (Please see the 2 swing sample pics below).


Even "25 fps, resize to 384x288" (8) is better.
(4) 25 fps
720x288
Discard field 1

74.2%

 
You definitely lose the half info, movie is unfluid, but result is sharp and saves you about 25%. Only 25%, I might add, because you give up 50% of resolution and 50% of fluidity.
(5) 25 fps
720x288
Discard field 2

75.2%

 
Interesting: In this movie all odd fields were harder to compress than even fields.
(6) 25 fps
720x288
Field Bob
(Method 4a)

 61.6%

Sharpen+Drop every second frame.
This is astonishing: Drop every second progressive frame should be theoretically the same as "Discard field2". But since I use filters like "Field Bob", that blurs a bit, it's not.
(7) 50 fps
384x288
Field Bob
(Method 4a)

60.9%

Sharpen+Resize down (decreased width until it matched 4:3).
In my view this is the method of choice, if size matters a bit, but not too much. You get a superfluid movie and the loss of visible information is not too high and you keep your movie 4:3.

This could be the deal for you if size matters a bit.

(8) 25 fps
384x288
Resize down to 384x288

36.7%

 

You size down to about 27% of the original film BUT the movie is kind of strangly unfluid/blurred. For 27% of the original quality you get 37% size.

In Virtualdub the deinterlace filter "Blend" is the same as "Resize down" except for the resizing. So you get exactly the same result, except the Blend'ed picture stays the same size and the downsized one is a little more blurred of course, because of the smaller image size.

This could be considered a typical looking movie downloaded from the internet. A bit unsharp and not fluid, but small.

(9) 25 fps
320x240
Discard field 1

29.6%

Resized down to 320x240 after discarding every second line.

In my view, it's a matter of taste whether you like resizing down (8) or discarding+resizing down (9). (8) gives you a blurred picture, (9) gives you a unfluid but sharp movie. Because it's quite sharp you can resize it down more. (How can a lower resolution be better?)

But I could agree that (3) and (8) can give you a good quality if the footage has a lot of static areas (like news broadcasting).

Once you get used to its unfluid play you will appreciate the sharper image quality.

I would say this is for sure the method of choice if you have an NTSC movie, because the frame rate of NTSC (ca. 30fps) makes the movie more fluent anyway.

And when I say unfluent for PAL systems (25fps), then this still far better compared to typical internet clips with 15 fps or 12 fps. Shame on you, you internet uploaders.

The difference between (8) and (9) stays clearly visible even if you resize to small frame sizes, even below 384x288:

 
Method (9) (Discard)

 
Method (8) or (3) (Blend or resize down). You could call these effects "Ghosts".

This deinterlacing method a very good and fast solution:

1) You work within Virtualdub only, thus very fast, no Avisynth needed.
2) Discarding is fast.
3) Resizing is fast, because you resize only the half info after discarding.
4) Compressing is fast because you compress a lower height/width.
5) Height/width ratio is correct
6) Since most of you will resize the final footage down anyway, discarding a field (=losing half of the resolution) isn't that bad.
7) Leaves you with absolutely no interlacing artefacts.

This could be the "good speed/quality/size ratio" deal for you.

(10) 25 fps
720x576
Duplicate Field 2

137.5%

 
This is the same as "Insert Field2 instead of Field1" thus doubling your height to keep the original ratio. You will get quality enhancement because the compression artefacts are much less visible. See all reasons and drawbacks for resizing up before encoding.
(11) 50 fps
720x540
Field Bob
(Method 4a)

152.8%

Resize up (increased height until it matched 4:3)+Sharpen
The same applies here as to (10). The 2 differences to above are: 50fps, resizing is bicubic instead of just duplicating.

I would definitely use (11) instead of (10), because for 15 percentage points more, your quality will be 4 times as good. The drawback is that your computer may be too slow to decode the movie. In that case you may use (17) or (18).


(11) Duplicate Field 1


(11),(17),(18) Discard Field 2 and resize by bicubic.

(12) 50 fps
720x540
Field Bob
(Method 4a)

158.6%

Sharpen+Resize up (increased height until it matched 4:3). So the same as above but sharpened first
This is just to show that the order of filters matters when compared to the one above. The result is pretty the same though.
(13) 50 fps
720x288
Field Bob
(Method 4a)

98.3%

Sharpen+a bit of the "Smart Smooth" filter
Don't do it. You lose a lot of details, compression is slow as hell (14 hours for 42 minutes film) and you don't gain much. If "saving filesize thru smoothing" is really needed, use DivX internal smoothing filters instead. They are optimized for to save filesize via smoothing.
(14) 25 fps
720x576
Smart Deinterlace

130.1%

Smart deinterlace Filter for Virtualdub by Donald Graft.
Options: "Frame-and-field differencing", "Compare color channels", "Use bicubic", "Motion threshold: 15", "Scene change threshold: 100"

Please note that this is a very good filter, that leaves you with a 25 fps (thus unfluent) movie though.

(15) 50 fps
720x576
Deinterlace Smooth
(Method 4b)

195.0%

"Deinterlace Smooth" Filter for Virtualdub by Gunnar Thalin.
If you are a quality fan then this could be the best quality deal for you. This filter is unbelievably good, but usually leaves you with twice the filesize (well, not always) and twice the height (=original height). This could make the movie unplayable (PC too slow). Besides the encoding itself is slow.

I love the results though. Brilliant, excellent, the way all video should look like: Fluent movements, sharp still scenes (backgrounds, unmoving objects).

To give you an idea, how big files are: I encode my DVB recorded MTV videos (528x576) with (15) and they are between 30MB and 170MB each. The filesize is mostly depending upon

a) How sharp is the source material (DivX loves blur).
b) Are there black letterboxes at the top and bottom? (Less to encode).
c) How many movements are there?
d) How dark is the overall video? (DivX loves darkness).

So the letterboxed, quite slow, unsharp, sometimes dark music clip

"Robbie Williams featuring Nicole Kidman - Somethin' stupid" (a cover of Frank Sinatra's hit)
is only 10MB per minute (= like uncompressed .wav)

while the non-letterboxed, many movements containing, quite sharp, bright music clip

"Red hot chili peppers - Aeroplane"
is 40MB per minute.

I faked Nicole's legs by the way. 

You may also see How Video Filesizes are affected by the Source.

(16) 50 fps
320x240
Deinterlace Smooth
(Method 4b)

45.2%

Like (15) but resized to 320x240
Unbelievable, that this resolution can give you such a sharp picture (but the encoding is quite slow).

Small size, sharp picture, fluent movements. Only 15 percentage points bigger than (9), but at least 4 times the quality.

This could be the quality/size/fluidity deal for you.

(17) 25 fps
720x540
Discard Field 1

99.6%

Discard and resize height up until width:height matched 4:3
The differences to method (10) are, that you resize by Bicubic Resizing instead of just duplicating the lines and that you resize to a smaller height than (10). This gives you a better quality as the pictures show (see (11)). AND: You get the correct aspect ratio. But it's still only 25 frames instead of 50, so I wouldn't use it.
(18) 25 fps
720x576
Discard Field 1

109.9%

Discard and resize up by bicubic resizing to double the height.
  The only difference to method (17) is, that you resize to the same width/height like in (10). This gives you a better picture but still results in a smaller filesize.
(19) 50 fps
720x576
Avisynth's internal BOB filter
(Method 4c)

177.6%

Use "bob" in the Avisynth script: Method 4c) on the Deinterlacing Homepage.
  Why use it? Nearly the same filesize as (15), but not near as good. Use (16) instead, if you want to decrease filesize effectively while still keeping very high quality.

This filter leaves you with twice the height (=original height). This could make the movie unplayable (PC too slow).

(20) 50 fps
720x576
Smart Bob

223.6%

"Smart Bob" filter for VirtualDub by Donald Graft
  Quite good filter. About twice as fast as (15) but a little worse quality and a bigger filesize. However in some cases this filter may produce smaller files than (15).

 

Now I wanted to check whether it makes a difference HOW I downsize the movie.

Method

%

Important Details

1. Resize down (bicubic)

100.0%

resized down to ca.49% by changing the width only
2. Resize down (bilinear)

97.8%

resized down to ca.49% by changing the width only
3. Resize down (nearest neighbour)

101.2%

resized down to ca.49% by changing the width only
4. Resize down/down (bicubic)

97.6%

resized down to ca.49% by changing the height AND width
1) Because resizing always blurs it is no wonder, that resizing height AND width (method 4) blurs more than width alone and thus results in a lower filesize since DivX gives you a smaller filesize the more the movie is washed out.

2) It is interesting, that the differences between the resizing methods of Virtualdub are nearly invisible (but they are visible). So I would suggest you use the bilinear method. Important: Bilinear resizing can give you strange effects with interlaced video. So deinterlace your footage first, when you decide to use it.

3) The best method for quality is bicubic resizing, which shows the same quality results, as if you resized by Adobe® Photoshop®.

 

How do the filesizes compare when using
different compressions = different DivX Quantizers
from the DivX Compression menu?

Setting the DivX Quantizer

As you can see from the graphics below from 0% to 80% the file size stays about the same, while the quality improves much more. But keep in mind that everything below 80% can hardly be called "good quality".

The filesize increases very fast as you add a few more % above 80%.

Above 95% quality doesn't increase too much, while the file size jumps up exorbitantly. So there's not much reason to go beyond 95%.

Below 90% quality drops faster than you save bytes. So there's not much reason to go below 90% either.

Stick to 90%-95%. Or if space is a big issue 80%-90%.

To see how quality is affected by the quantizer see my DivX Quality vs. Quantizer examples

 

Conclusions

1) Compressing movies with 25fps instead of 50fps does not halve the filesize to 50% (but to 75%).
2) Smoothing filters produce a greater loss of quality than they save size, except for material with much noise (=snow).
3) Setting quality below 80% results in a low quality movie without saving much size.
4) To decrease filesize decrease (bilinear) width/height (after deinterlacing by discarding), use less keyframes and play around with the quantizers between 80%-95% quality.

 

100fps: How to deinterlace Video the right way. Examples, Tricks, Comparisons