let's figure out what 24f actually is...

[John Cordell]

Over in HVX200-land, they had the "what's the CCD rez" question for quite a while, with many cool and interesting and clever approaches to deducing the the answer. And now they've got the "what's the true effective chroma resolution due to pixel shifting?" issue, which I also quite enjoy. Both of these issues have created discussions that have taught me a lot.

So the big question here in XL-H1 land is, what is 24f? Simple de-interlaced 48i? or some smarter motion adaptive version? Has anyone thought of or done any experiments that might lead us to an answer?

[Robert Sanders]

Unless someone can get inside the DigicII software processing, I'm not sure we can find out. But I suspect it's some straight forward deinterlacing of 48i.

[Thomas Smet]

This test should be easy to at least start to get some answers. Somebody with a H1 needs to shoot something with slight moving thin details at a very low angle. For example some power or telephone lines outside. Here we will be able to tell what is exactly happening on each line to the data.

From what I have seen so far 24f looks amazing but I cannot tell what the heck is going on in the 24f footage.

[Thomas Smet]

There is an image I have recently looked at from a Cineform test between the H1 and the Viper. There is a shot captured from SDI and the H1 of a car outside and there are some power lines in the background.

Whatever Canon is doing I do not think it is a straight deinterlace. The power lines curve up at a slight angle and all through the length of the line there are no missing chunks or aliased edges. Even with an interpolation during deinterlacing you will have problems with objects or details that are one pixel thick. For example say the power line runs through line 3. During deinterlacing line 3 happens to be on field 2 so line 3 either gets dropped and becomes a duplicate of line 2 or an blend of line 2 and line 4. Since the power line wasn't on line 2 or line 4 this will make the power line not show up for that segment.

[Barlow Elton]

Definitely not a straight deinterlace--it's much more sophisticated than that. I think it does something motion adaptive, because the results remind me of the Compressor conversions of 1080i to 720 60p in that it does not look to have any temporal displacement whatsoever and when you step through the frames--everything looks perfectly progressive. If it were a straight deinterlace, the resolution wouldn't hold up the way it does in 24F. If you do a simple deinterlace 50i to 25fps there is an obvious drop in resolution...Cineframe 25 mode on the Z1U being a prime example.

Not sure that gets us any closer to a true answer but I am very confident it isn't a simple deinterlace of 48i. There is a noticeably slight loss of sensitivity when flipping the switch from 60i to 24F...which would seem to support the 48i/hz theory.

My belief: There's something else going on, and maybe it's something Canon is reluctant to disclose because it's likely a proprietary process that they don't want their competitors to understand. It may be a combination of a number of things (deinterlace, reclocked CCD's) but the proof is in the pudding...and who can honestly say that 24F doesn't look progressive?

[Greg Boston]

I have been giving some thought to the magic of 24/30F. Here is what I have hypothesized.

The CCDs are being clocked odd/even at a very high frequency. So high, that motion can't easily be perceived between the two clock pulses. Then it spits out 24 or 30 of these per second.

_M__M__M__M__M__M__M__M__M__M__M__M__M__M__M__M__M__M__M__M__M__M__M__M

Imagine the letter M is representing two quick clock pulses that read the odd/even lines off the ccd block.

I may be all wet with this theory and there is likely more getting done via the DigicII chip, but it's kind of fun to speculate about it.

-gb-

[Barlow Elton]

Greg,

Forgive my simplistic questions but...would this higher clocking of the CCD affect motion blur at all? The blur of 1/48th in 24F mode looks exactly right.

Would this theory of yours mean the camera basically jumps into "hyperspace" or warp speed to pull some tricks, and then comes back to what the human eye can actually perceive as properly rendered 24p?

I like the speculation too...but it would be nice to at least have a layman's understanding of how Canon is pulling it off, just for the academic interest. All I know right now is they are definitely doing something right in 24F mode.

[John Cordell]

Greg, I had wondered what I think is the same thing: namely, what's the fastest that the two interlaced fields can be extracted from the CCD? If it's quite fast, it would make sense to grab the two fields as close in time as possible so that the time difference between them is small. I also wonder if there's a way to stop them from accumulating, in which case you could grab the two fields at your leisure, at different points in time, but they still represent the same moment in time. Resolution chart testing seems to have indicate a loss in vertical resolution when going from 60i to 24f modes, so whatever is going on likely isn't completely the moral equivalent of progressive scan.

Any CCD experts' opinions on possible trickery at that level would make for interesting reading for me.My problem is that I don't actually know how interlaced CCDs work vs. progressive, so there is little signal and much noise in my random musing on this subject.

[Peter Moore]

It would indeed screw up the motion blur if the sampling was faster than 1/24. Shutter speed would be meaningless then.

[Thomas Smet]

I think we should look at the chroma channels to see what is going on. A lot of time the luma can be interpolated or pixel shifted in many ways to make it seem like you are getting more resolution. I do not have enough decent images from the net to figure out what the chroma is doing. I took a look at some of the images Cinform has done from SDI but I cannot make anything out yet.

One area I have been thinking of is maybe they are using the same pixel shift trick as Panasonic. The Panasonic only starts with 540 vertical pixels just like a single field from the H1. With pixel shift they could in theory get more resolution than 540 but not near as much as 1080. This could explain why there is a drop in resolution in the F modes compared to interlaced. The only problem with this theory is that I would still think there would be alias issues due to one field having only every other line. I'm not sure how an interlaced 1080i chip could perform as a perfect 540 progressive chip. I would really love to find out but at the same time if many of us our inspecting the footage with a microscope and cannot seem to come up with any artifacts does it really matter. I want to know just because I am a dork and it would bug the heck out of me by not knowing what my camera is doing.

One other point. I think this test should really be done with SDI. Interlace and progressive mpeg2 handles chroma in a different way and gets very hard to compare based on that. SDI should also show us what is happening in the camera to get F resolutions without any other issues thrown in.

[Barlow Elton]

Originally Posted by Thomas Smet

I want to know just because I am a dork and it would bug the heck out of me by not knowing what my camera is doing.

Amen. Dorks wanna know!!

[Greg Boston]

Originally Posted by Barlow Elton

Amen. Dorks wanna know!!

The reason I spent a quarter century of my life as a technician is because I couldn't stand not knowing how things work. And if you know how it works, you can usually fix it when it doesn't work. Videography is actually my second career.

-gb-