Which is better: 24F or Magic Bullet?

[David Ziegelheim]

Undocumented 24F takes advantage of the various features in the camera. Magic Bullet claims to "unique combination of field based pattern matching and motion compensated deinterlacing techniques delivering the sharpest results on the market".

Which is better?

[Lauri Kettunen]

In the European Broadcasting Union they've had a long debate of the HD standards, and the question of compensating deinterlacing is one of the issues. Technical people do seem to agree that none of the deinterlacing compensation techniques are fully satisfactory.

As a rule of thumb one could argue that the best choice is a progressive CCD, then the second best alternative is a solution embedded in the camera such as Canon's 24F, and the post processing strategy is the one thereafter. This argumentation relies on the assumption/fact the there is more information available in the camera than what one can gain from the image produced by the camera.

[David Ziegelheim]

Although, pre-compression, the camera has more information, it is also processing in real-time and would be dependent on the algorithm. I thought for the Sony's, post-processing was preferred. In the Canon, the CCD has higher resolution, however the 24F reduces the resolution significantly. Is there a clear answer for this camera?

[Chris Hurd]

No it is not reduced "significantly." Have you looked at any of the sample clips provided here on an HDTV?

[David Ziegelheim]

I believe the reviews have indicated it is about 540 line resolution, vs 1080 interfaced and 700 or so for the HD100. The 540 lines I believe was similar to the HVX200.

I thought the BBC always used software for a 24p conversion, and the Cineform site references Magic Bullet and DVFilm Maker several times. Has someone done tests with post-processing software vs the Canon 24F mode?

[Chris Hurd]

Have you looked at any of the sample clips provided here on an HDTV?

[David Ziegelheim]

I've looked at the sames in the two shootouts on my 1600x1200 monitors. With a lot of those images, it was hard to tell from the review whether it was 60i or 24f, and whether it was from the HDV tape of HD-SDI output.

Do you know of some definitive images for comparison? What are we looking at in this-- http://i.cmpnet.com/dv/magazine/2006...TX-64-XLH1.jpg -- 1920x1080 image?

[Thomas Smet]

that is a 60i image. You can see the interlacing in the girls hair.

The only problem with deinterlacing 60i in post in the fact that 1080i video is filtered to about 810 lines. I do not mean there is a 1920x810 image but the amount of detail is equal to that of 810 lines. If you deinterlace this each field actually only has 405 lines of detail compared to 540 from Cineframe and the any where between 540 and 810 lines for 24F. Unless you use a tool that is very very slow to try and rebuild the detail you are only getting 405 lines of detail per field. This isn't really such a bad thing but it does show that even if 24F or 30F is only getting 540 lines (which I think it is getting more depending on the color of the scene) it is still better than 405 so maybe it isn't such a bad thing.

The other thing you have to think about is aliasing artifacts. With a normal deinterlacer you will end up with odd edge artifacts, shimmering details and thin lines that flash on and off. 24F and 30F seem to not have these issues. While the resolution may be lower it doesn't look like deinterlaced video. There are no thin object artifacts because any of those problems are not captured in the first place in the camera. You will end up with a very clean overall pleasant image no matter what level of detail or motion is in the scene.

Something else to think about is the fact that 24F and 30F use a true progressive mpeg2 encoding with a chroma pixel block of 2x2 pixels. Interlaced mpeg2 has a really messed up form of 4:2:0 where each field has a chroma with a kind of a 2x4 pixel block. The chroma samples alternate on the fields which can make a huge mess. When you discard a field with interlaced mpeg2 video the chroma really suffers. Using 24F and 30F will give you a much more natural form of 4:2:0. Since 4:2:0 duplicates the chroma every other line 24F HDV doesn't really loose any chroma detail compared to 60i anyways. Both formats would loose the same amount of chroma. The only thing you really gain with 60i HDV is luma detail.

[David Ziegelheim]

Was the image captured from HDV tape or directly with HD-SDI?

I thought that Canon stored the image with a 2:3:3:2 pulldown. Doesn't that waste a lot of the bandwidth?

Magic Bullet and DVFilm Maker both claim to have algorithms that look at all of the lines to synthesize the missing information. Howerver, I don't have any experience with them.

That image shows much more detail than the 1280x720 HD100 image in the attached blowup. If vertical resolution is only 810 lines, and the HD100 is around 700 lines, why is the difference so large? Was it HD-SDI vs analog component HD? Was it compression algorithms to tape? Was it a different focus (the HD100 image seemed to have leaves behind the subject in focus, that were out of focus on the H1)?

[Thomas Smet]

Originally Posted by David Ziegelheim

Was the image captured from HDV tape or directly with HD-SDI?

I thought that Canon stored the image with a 2:3:3:2 pulldown. Doesn't that waste a lot of the bandwidth?

Magic Bullet and DVFilm Maker both claim to have algorithms that look at all of the lines to synthesize the missing information. Howerver, I don't have any experience with them.

That image shows much more detail than the 1280x720 HD100 image in the attached blowup. If vertical resolution is only 810 lines, and the HD100 is around 700 lines, why is the difference so large? Was it HD-SDI vs analog component HD? Was it compression algorithms to tape? Was it a different focus (the HD100 image seemed to have leaves behind the subject in focus, that were out of focus on the H1)?

No the pulldown doesn't waste bandwidth at all. In fact it is the only 1080i HDV format that saves bandwidth. The pulldown is added during output. On the tape it is a true 24 fps mpeg2 video. That is why so many NLE's have problems with the format. Canon went with the method of mpeg2 that gives the highest quality but is a very special form of mpeg2. The pulldown is just duplicate flags that do not take any bits. Therefore you have that many more bits per frame. SONY wastes bandwidth by putting their 24 fps format with the pulldown built in into a 60i video. Even though many fields are duplicates the encoder and decoder treat them as whole new fields wasting bits to encode the same thing twice. Since the 60i encoder doesn't even know some of the fields are duplicated it treats everything as a whole new frame and encodes it like that.

A very good (but slow) method such as DVFilm Maker can do a fairly good job at trying to make up the missing detail but it isn't perfect. Even if it does a very good job it is very slow. It is much easier and faster to just deal with the 24F and edit in a 24p project and be able to output right to your final format as 24p.

One other advantage to using 24F is the fact that you are shooting in a 24fps and can tell if a shot is working while you are shooting. If you shoot at 60i a shot may seem to work but when you convert it to 24p it may not work exactly the way you wanted it to. By shooting with 24F you know exactly how the motion will turn out and get the exact look you are looking for ir terms of pacing and motion smoothness.

About the 810 vs 700 lines. The reason is because you are looking at objects of two different sizes. You have to either scale up the 720p video to match or scale down the 1080i to match. You will see that they are fairly close when you do this. While there are only 810 lines of detail it is still sitting inside of 1080 lines so it can look like there is more detail there because the lines get blended together. Resolution isn't everything either. There are a lot of other facets that make up video quality. Take the sample images that you are looking at. Every image in those tests except for the JVC images are from a 1080i camera. While those images all share 1920x1080 pixels you can tell that some have more detail than others. A resolution test isn't going to test how many vertical pixels are in a image but the amount of detail it can resolve. How small can an object be before it starts to get blended into a blob of pixels. HDV cameras have many issues that lower the true level of detail. That is why for the most part video from the JVC camera can hold up to the level of detail from a 1080i camera. The JVC may use less pixels but it makes better use of a higher percentage of those pixels. 1080i on the other hand may waste a lot of those pixels. I say may because it is a lot more complicated than that. While 1080i may not show 1920x1080 unique pixels it does show a blended version of ????x???? which may make up a smooth image that can give the illusion of having a lot of detail. For example on a 720p image a tree turnk may be 4 pixels wide. on a 1080i image that same tree trunk may be 6 pixels wide. While those 6 pixels may be a blended version of the 4 pixels it still looks like the tree has more pixels. the 720 blown up to 1080 depends on math to interpolate the in between pixels. The 1080 uses physics and laws of nature to come up with the in between detail which to some people may give a more natural realistic look even though it may not have more detail.

[David Ziegelheim]

Those images were different sizes because the they had different numbers of pixels. The detail in the black strip and the shades in the lavender strip are clearly present in the H1 image and not in the HD100.

What was unclear was if the image was recorded to tape, and which lens was used.

Does the Canon actually use that bandwidth when it marks the frames duplicates? How is that done...or is that subject for a different forum?

I image this would affect the XH-A1/G1 and HDV10 also.

[Christopher Glaeser]

Originally Posted by David Ziegelheim

Does the Canon actually use that bandwidth when it marks the frames duplicates?

The duplicates are not recorded to tape.

Originally Posted by David Ziegelheim

How is that done.

If two frames are identical, omit one and mark the other.

Best,
Christopher

[David Ziegelheim]

I probably didn't ask the question well. Does it change the compression algorithm to use the extra bandwidth? In order to use the bandwidth, wouldn't the image have to be stored differently?

Are there screen shots on line of 24F saved to tape and 60i saved to tape? With an HD100 for reference?

Thanks,

David

[Chris Hurd]

For those outdoor shots, all of that was recorded to tape.

The HD100 was not a "reference" camera; it was one of several HD camcorders that we were comparing in side-by-side tests for the Texas HD Shootout back in April. Our reference camera was a Panasonic AG-HDC27H VariCam, but we did not take it to the river that day for these particular frame grabs that you're referring to above. Cameras present for that sequence of shots were the Canon XL H1 with 20x HD lens, Panasonic AG-HVX200, JVC GY-HD100A, Sony HVR-Z1U, and Sony PDW-F350 with a Fujinon 2/3rd-inch adapter and lens.

[David Ziegelheim]

Are any of the Canon images 24F?