4:4:4 12-bit Uncompressed DVX100

[Robert Martens]

Downloaded that PSD file, opened it in combustion (I don't have PS), and tinkered a bit. A little gain boost, Median noise filter with a .5 radius--completely eliminating the noise--saved the result, cropped and resized it, and boom! Looks pretty good by my standards. The JPEG compression artifacts help alleviate the smoothing done by the Median operator, and aren't too ugly.

You sure ain't kidding about the details in there. :)

I'd show off my result, but the FTP password for my website that I've been using for the past few months no longer works, courtesy of the geniuses at CI Host (the same people who changed my email password a month ago without telling me). But that's a whole other post.

Obin, I can't seem to access the URL you posted for the "juancap_colorwork" image. The other two show up just fine, but that one gives me a "Page cannot be found" error.

[Stephen van Vuuren]

Obin:

I can't get all your links to work. And do you know how to change them to be clickable? If not, let me know or search here for the code.

[Juan P. Pertierra]

Robert,

Email me the image and i will post it on my website. I just got extra space added to it so now I can post some clips when i'm done with all the work i have this week.

The link to one of Obin's file that doesn't work is because my name is spelled wrong :) thus it is:

http://www.dv3productions.com/upload..._colorwork.jpg

[Robert Martens]

It's on its way, Juan, should be there soon.

I've contacted tech support for my website, and usually hear back within a day or so; as soon as I'm able to log in, I'll take the job of hosting my file off your hands.

Thanks a bundle!

[Robert Martens]

Talk about fast response -- got the correct password, uploaded the file, you can view my image here:

http://www.gyroshot.com/images/juancombustion.jpg

Update -- I've uploaded an additional (larger, better looking) version of the image; since everyone else seems comfortable with uncompressed TIFFs, I figure what the heck? You can still access the JPG, if you like.

http://www.gyroshot.com/images/juancombustion.tif

I would also like to know what's with the green and magenta edges on everything in the source photo...any ideas? It's especially noticeable on the sidewalk, as it passes behind the vertical bars of the guardrail. Does this have something to do with the alignment of the color channels, perhaps?

[Juan P. Pertierra]

Here is my somewhat 'corrected' version, i couldn't get rid of all the speckles using photoshop only...

http://expert.cc.purdue.edu/~pertierr/cap5_RAW.tif

[Obin Olson]

Juan, what do you think? bad pixels? or?

[Brett Erskine]

This is why past my realm of knowledge but when I mentioned this pixel problem to someone much more experienced he said that manufactures are allowed to have a certain number of dead pixels in a display when they create them. Apparently they have the means of masking these dead pixels from showing up but since your pulling off raw files off the CCD you might be by passing this masking and revealing the dead pixels.

Again I dont know if this is correct. Im only passing on the info that was told to me. Im not even sure techically what they are doing to mask the pixels and whether that masking would or wouldnt come into play when you are doing raw files. Any thoughts?

[Obin Olson]

i would say yes BUT the pixels are moving around in each frame...a dead pixel stays dead in every shot....

[Dan Vance]

Juan,

I've tried to read through all the posts to come up to speed on this thread, so maybe I've
missed something that's already been said, but...
I'm not sure you're going to be happy with your results when you get where you're going,
because:

You're starting with 4:1:1 and trying to get 4:4:4. But that first 4, which you already have,
represents luminance samples, and that's where, practically, all your "resolution" is. So,
you can expect to get significantly better color rendition, but NOT better resolution. The
resolution of your Y output (of the Y/C output), is the full resolution that the camera is
capable of. There is essentially no resolution loss in the Y signal going from the chips to
the Y output. That will be a little better than the equivalent compressed DV luminance,
but not a whole lot. And of course you can get at that without going inside the camera.
Also, at one point you said you're limited "only by the glass and CCD size"!! Yes! But the
overall performance can only be as good as the worst component. The DVX100 lens
probably has a resolution of 500 lines or so, or at the very best, 600 lines. And who knows
what it does to color? It's good glass FOR THE MONEY, but relative to pro lenses, it's
pretty mediocre. So even if the CCD was capable of 1000 lines, the lens limits you to 500.
You're improving the milkshake, but trying to drink it through the same tiny straw.
You also seem to be neglecting the bandwidth limitation of the NTSC signal, which is 540
lines, and there's no way around that, as long as the signal is NTSC.

So, you're doing a lot of work for a little more color info. Is it worth it? Maybe. I certainly
would be the last person to dissuade anyone from experimenting. But the physics say it
might not do what you're expecting.

Also to clarify some other posted info, PAL CCDs and NTSC CCDs are not
interchangeable, because the pixel aspect ratios are different. And "5:1" compression does
NOT in any way "throw away 80% of the information." Compression is not truncation,
it's an analytical processing method of getting rid of as much UNNEEDED info as
possible.

As you've mentioned, the Canon XL1 is another candidate, and at least you could put
better glass on that one, but unfortunately, it doesn't do 24P.
I don't mean to be throwing cold water on this, and I think you should continue and see
how good you can make it, but just be aware of the limitations of the camera and don't get
your hopes up too high.

[Luis Caffesse]

Actually, you bring up some good points.

But what about the perceived difference in dynamic range?

Juans modification may not be increasing resolution, but getting a 12 bit signal vs the compressed 8bit signal should count for something, shouldn't it?

-Luis

[Juan P. Pertierra]

Dan,

Thanks for your input. However, i think there are some significant innacuracies in your post...

1.)I am NOT starting with 4:1:1 and going to 4:4:4. I am literally capturing 4:4:4 straight out of the CCD's, at the full sampling rate of the A/D's.

2.)You're right in that the lens on the DVX might do 600 lines at the most. However, afaik that doesn't really pose a limitation since the CCD's have 494 lines.

3.)I AM getting a little more resolution because of the fact that I am getting the full CCD frame. It's not much but if there is something i have learned so far is that the more information you start with before correcting your video for a specific look, the better.

4.)I'm not really sure why you're talking about NTSC signals, because I am not even dealing with that at all. What I am getting is a non-standard frame size, and it's up to the person working with the footage to decide what medium it is going on. If he/she wants to print the entire 773x494 frame to film, there is no NTSC standard compliance involved here ~anywhere~.

To top it off, i think a lot of people will agree that the very significant increase in dynamic range, as well as the fact that the 12-bit RAW output can display at least 48 times as many colors as the 8-bit DV output alone are worth it.

Finally, i recall some people doing up-rezzing experiments with the standard DVX output and getting results comparable(with better color) than the JVC single chip HD cam. I am almost 100% sure that uprezzing of this raw full color footage will yield even better results while we wait for a 3-chip HD camera to apply this on.

Oh yeah, and NO COMPRESSION ARTIFACTS. Hate those big blocks on the reds, blues, and have you ever tried pulling a chroma key? You can say that the DV compression only gets rid of uneeded data, but i think very few people here will beleive you.

Cheers!
Juan

[Peter Plevritis]

Just take a look at the images.

They are very impressive. And I believe these are only 10bit instead of 12.

The dynamic range is outstanding. The added resolution is a great plus. All great for going out to film; if that is what you want.

Hopefully Juan can eliminate the digital specks and move on to getting this thing production ready. (at least I hope he plans on that).

[John Cabrera]

Juan,

I think what Dan meant was that your starting with a camera that's already capible of 4:1:1 and your trying to get 4:4:4 out of it (just to point out that the 4 is already there for you in luminance and make his point about resolution) . I don't think he meant that you're trying to turn a 4:1:1 signal into a 4:4:4 signal.

Your other counterpoints do make sense, though.

Keep going with this... I monitor this posting every day. You're a true innovator, and asside from what this could do for indy filmmakers, the ongoing story itself makes for great drama. :)

John

[Thomas Smet]

I just took a look at the U and V channels of the latest sample image and we are clearly getting real raw 4:4:4 frames instead of an upsampled 4:1:1 frame. The seperate color channels are crisp, clean, and acurate. If you work with high end visual effects this is the ultimate feature I would want for this mod and so far it is working better than I expected. The dvx100 might not be better than a 1/2 or 2/3 inch camera but I would take 4:4:4 frames any day over 4:1:1 or even 4:2:2 frames. If you want to check out how much better 4:4:4 is then in photoshop change your color mode to lab color. The a and b channels are the U and V color channels and l is the y channel. Compare a raw frame to any dv frame and you will see the clear differance.