4:4:4 HD Uncompressed via Picture mode ala Juan's mod?

[Rob Hester]

disclaimer: I really don't know enough about what i am talking about.

There are megapixel CCD's on the cheaper miniDV camcorders, as well as the pxd10. If we could sample the image right off of this CCD, would we be able to capture @ a higher res?

My thoughts against this idea:
I know some cameras have pixel shifting which samples a higher resolution by taking 3 snapshots and reconstructs the image as one (is this really an over simplification?)

I also know that video pixels and picture pixels hold different ratings on camcorders....however i don't exactly know why- so should this be a reason to assume we couldn't obtain this resolution?

My thoughts for this idea:
On my camera, when I switch to d.wide mode, the picture on the lcd gets choppy. After testing, I know that this is the difference between d.wide being 30P and normal mode being 60i, so the motion signatures are different.
BUT what is interesting, is that there is absolutely no difference between the motion signature in picture mode, and d.wide mode - but the difference from normal mode to picture mode is obvious. So what I might conclude (although a huuuge stretch) from this is that my camera's CCD can capture a maximum of 30P. This follows, because the d.wide mode has better resolution than the normal mode - signifiying to me that the d.wide image is downcoverted into normal DV resolution (although I have no idea how much). To support this, I know that in d.wide, the EIS is disabled, so at the very least, the increased resolution is from the extra CCD space usually reserved for EIS.

What i am wondering, is how do every one else's picture modes opperate? Is this a possibility at all? MY intuitions tell me it really isn't possible to get 1024x768 video resolution out of a cheap 1 ccd camera....I just don't know enough about it to know why.

If the camera can display the image in the LCD for me at 30P BEFORE i press that snapshot button, couldn't a mode ala Juan's DVX100 try to capture the raw flowing information?

Thanks for your time
Rob

[Rob Hester]

alright, i guess there really isn't any interest in this...sorry to clutter the board!
thanks to the 96 who read it though! ;)

Rob

[Rob Lohman]

It's certainly interesting, BUT, I think most people are waiting to
see what happens with the DVX thread. I'm certainly interested
in all of this. The other problem is that there usually are only a
handfull of people who have the knowledge to do such things
*AND* want to trisk their camera.

[Wayne Morellini]

Please keep going, we defintely need this Rob.

With 35mm lense adaptor, GS-70 (or gs 400) and a small hardrive we would have a sensational hobby cinematic camera. If we got really ambituose, and physically hack the camera, the lense could be replaced with something a bit more directly, for more light. With HD we can go to a full inde camera. If the firmware is changed the full 4:4:4 frame could be sent out the firewire port (on 200-400M/bs+ uncompressed for SD),artifical gain and lux constrictions be removed, and with HD, the comrpessor reprogrammed to loosesless compression to fit through the firewire port.

I suggest to go to the other thread and make a post letting them know that your thread is here if they are interested.

The biggest problem though, with getting HD from a mega pixel camera, is that the frame rate might be restricted by the data rate of the CCD, or the CCD might not be able to physically process it fast enough. The manufacturers probably deliberately leave these design restrictions in, but there maybe a couple of cameras out there that don't have these restrictions. If you discover these cameras, then I think there will be some potential indie people screaming their heads off in enthusiasm ;)

[Rob Hester]

Okay, you guys know how I was talking about my d.wide mode? IT says .7x free...if you need me to explain better i can...i am in a rush because I have 2 exams tomorrow!!!

Anyways, I resized the downsampled d.wide image to normal mode with the same picture as a reference, and upon expanding the d.wide to match the normal, I got a much higher original resolution for d.wide! The d.wide has a resolution of 962x641...now if i could ONLY get it to not resize down to 720x480!

Here is the screenshot: (you need to right click and "save as")
http://www.aloofhosting.com/robh/extrares.jpg

Sorry about the rush, wish me luck on my exams!
Rob

[Rob Lohman]

Keep in mind that the firewire 400 port does 400 MBIT/s (or mbps),
NOT MB/s. So it roughly translates to a max of 40 MB/s which you
can quickly fill up uncompressed. It won't allow for HD resolutions.

For example, here is a small table with throughput rate if we
want to capture full RGB channels @ 10 bit per channel (which
yields 30 bits which takes 32 bits to store). Calculation:

Horizontal res x Vertical res x bytes per color x fps

or

? x ? x 4 (=32 bites) x 30 fps:

720 x 480 x 4 x 30 = 39.5 MB/s (31.6 @ 24 fps)
1024 x 768 x 4 x 30 = 90 MB/s (72 @ 24 fps)
1600 x 1200 x 4 x 30 = 219.7 MB/s (175.8 @ 24 fps)

You can see how quickly it goes. Firewire 800 (which is not in
use much) has a theoretical limit of around 80 MB/s. I'm not sure
which current implementations can even get there. And that won't
even get us up to 1024 x 768 (unless you drop down to 24 fps).
So if you want to go to HD then either use something else then
HD or go with some form of compression.

Now this doesn't have to be lossy compression (like DV or MJPEG
or MPEG), but can be something lossless like HUFFYUV for example.

Anyway, datarates start to increase dramatically FAST. Ofcourse
storing 40 MB/s is going to be a tough thing to do. Storing 90
MB/s is only possible with a RAID set I think and I'm not quite
sure if we can get to 219 MB/s just yet.... That's a LOT of data.

Just to give everyone an idea....

[Rob Hester]

Yeah, this definately would be a problem. JVC used mpeg2 extremly compressed to fit into dv25. My initial reaction to that was that, well, mpeg2 is an old, outdated codec. There exists many more efficient alternatives if huffy (there is another one as well) is still too much data to throw around. Divx, xvid etc...as long as they can do HD. One in particular, to me, stands above. Windows Media 9. Obviously we would like to have uncompressed images to edit with, but if that isn't possible...we'd want the absolute best compression with the least amout of video artifacts. I was snooping around for WM9 HD video rates, and here is what I came across (check page 14 @ http://www.avs.org.cn/avsdoc/2003-7-...F%E6%BB%D4.pdf)
______________________________________
1080i/p - 3-20+ Mb/s
720p - 4-6 Mb/s
On page 15, they identify sweet spots for compression. They list 5-8 Mb/sec for DVD, and 18-22Mb/sec for HD.
______________________________________
I see no reason for the codec to have an artificial ceiling, so, we could probably pump 40Mb/sec if we wanted to, even more...(does sony do something like 72Mb/sec for HDCAM? I know nothing about it, I think I read that the other day though). My 120g Western Digital Hard Drive does 33MB/sec (mega bytes, not bits) I believe...so that is more than enough. The drive has an 8meg cache...but RAID, or what have you and throwing 72Mb/sec through hard drive shouldn't be a problem.

Here are the problems as I see them:
-Running out of disk space VERY QUICKLY
-Having a processor capable of encoding WMV HD which is also somewhat portable (I searched for required encoding specs, couldn't find any...I know my 2.66Ghz P4 can do it...but we are talking realtime here...and this would probably be a big issue)

Also, now that we are getting really crazy...we could think about encoding the audio using WMA9 Pro Audio (24bit/96Khz sampling) and can do 5.1 surround sound encoding too! Okay, now I am really getting up in the clouds...I realize that, but there are solutions out there which, although not perfect, may still suit our needs.
Anyone have any input on the higher resolution of the d.wide mode? That's a promising sign that other cameras with EIS might have more resolution as well! I would really like to know how much a pdx10 would have.

Anyways, back to studying...I'll probably take another break in a few hours :S
Rob

[Rob Lohman]

I highly highly doubt any current processor can do realtime WM9
encoding with their codecs. Why? Because you already need a
pretty kick-ass PC to playback their HD stuff. I have a feeling we
first need to stick with SD to make this all possible.

The easiest way to use some form of compression is to get a small
PC in there. There are several standard PC's that come in very
small form factors and I think there are also certain embedded
systems that are full PC's at a very small size (which you can
expand, hardware wise, to what you need). So standard it has
almost nothing on it's main board.

But then again, the camera would be the first step....

[Rob Hester]

Scroll down for image capture

I spent a lot of my day looking for PCI HD encoders and to my dismay, all were VERY expensive. Any one have any insight on solving this problem?

Rob, I think you'll like this post :D. I have your "Letterbox Caculator for IE" (http://www.visuar.com/letterbox/calc.htm) site bookmarked and it came in VERY handy for my test. I did another test with my camera to get a more solid number for resolution of the CCD. In this test I compared the normal mode, squeeze mode, and d.wide modes but with the main focus of finding the actual resolution of the d.wide mode before downsizing. I used Rob's site to caculate how to crop the d.wide image down to 16:9 and most interestingly, the vertical resolution would still be 482! So if only I could tap into this CCD, I could a) have a new anamorphic squeeze mode for SD MiniDV, or, b) i'd get a 16:9 480P image! This wouldn't be as hard to encode as HD, and (oh boy my head is spinning now) if I had the brain of Juan, it might be possible to bypass the miniDV encoding, and encode 480P on miniDV tape at 25Mb/sec with Divx, Mpeg4, or WM9. Even Mpeg2 at 25Mb/sec is much better than DV...if anyone has that older site which compares codecs, there is an mpeg2 one thrown into the batch.
---------------------------------------------------
Here is my image, showing the differences between all 4 modes with outlines to separate the image area of each respective mode.
http://www.aloofhosting.com/robh/480P.jpg
(Right click, "save as")
---------------------------------------------------
Findings:
Normal: 720x480
Squeeze: 720x360 - this mode chops off the top and bottom, and then squeezes the vertical resolution back up to 480pixels. This leaves irreversible sharpening artifacts.
D.wide: 954x637
D.wide cropped/16:9: 954x482

[Rob Lohman]

Glad it was of use. I edited your link since it was linking to the
old page! I have since then regained access to my ftp server and
the thing is now up in a more permanent place:

http://www.visuar.com/letterbox/calc.htm

Please use that link in your favorites, thanks.

I'm not sure what you are getting at or doing here Rob. Which
camera are you using now?

What aspect ration did you use in my calculator. You should
probably use 1.0 since it's a still camera, right?

[Rob Hester]

I'm not sure what you are getting at or doing here Rob. Which camera are you using now?

This is my JVC DV3000U in 3 different video modes (miniDV) so 0.9 pixel aspect ratio is correct and the resolution is 720x480. However, if the camera is put in a static location, and I capture from each of the modes the d.wide mode records a much larger area to tape. The camera says it is like getting a free 0.7x wideangle lens, but it is really just sampling the full CCD (it uses the extra CCD resolution that is normally used for EIS).

On my camera,
Normal mode doesn't use the full CCD and reserves some space for EIS use.
Squeeze mode doesn't squeeze anything, instead it chops off the top and bottom of the normal mode image, and then stretches the vertical back out to 480 pixels @ 1.2 pixel aspect ratio. "Squeeze" is technically an incorrect label for this mode...i'd call it stretch-and-create-over-shapening-mode! For the purpose of this test, I unstretched the vertical.
D.wide mode uses the full ccd, and records to tape at 30P, not 60i. What i wrote earlier, was that it's interesting to note that although the picture mode on my camcorder (which goes up to 1280x960) produces a different picture (shows more height and width spacially) from a normal-video-mode captured at DV resolution (720x480), when d.wide-video-mode is captured to tape, it actually recreates the same spacial height and width that the picture mode would. This led me to believe that the d.wide was being captured higher and then downsampled to 720x480. So, I took two images (initially), one from normal, one from d.wide, into photoshop, and scaled the d.wide back to the proportions of normal (if the camera was in the same position when shooting both, then the apple would appear different sizes between the two modes--smaller in relation to the entire image when captured in d.wide). Since i know there is no optical difference, I know it is the CCD downsizing, and so re-scaling the d.wide image would recreate the picture as it was captured before being downsized. So, this gives me the approximate resolution of d.wide before downconversion. You can do the same thing if you place the normal mode image inside of the d.wide and scale the normal down and you get the same effect.

The point of this thread, was thinking that people could use picture modes on their camera to get nearer HD res. My tests show that this is the highest resolution that I am sure 30P can be captured at, just as Juan has done and found that the raw image was 771x494.
John Gaspain, on Juan's thread also said
"I will also get higher Resolution, 934x768 is the actual CCD resolution, because the cam down converts to 720x480 to fit on DV tape."

I have made this WAY more complicated than it should have been...if anyone wants to recreate my test, I can host a second of raw DV from each of the 3 modes for you? Rob, you seem to be the only one really interested...so I hope i've cleared this up for you

Thanks for your time,
Rob Hester

[Rob Lohman]

I think you did. I'm just not sure where we are heading with all
of this. We need some more technical people in here!

[Juan P. Pertierra]

Most of the camcorders(like the Sony's) that have CCD's for 1MP+ stills, cannot sample fast enough to make fluid video at the high resolution.

That is, they only use a smaller area of the CCD for video and then they use the entire CCD for stills, sampled more slowly.

There a LOT of cameras which the CCD's record full-frame video and are larger than 720x480. However, i haven't found any 3CCD camera under $4000 which has an incredible amount of extra untapped video resolution, but I might be wrong.

There are 1chippers which have larger CCD's for video, but this is because of the mosaic pattern applied such that the color is only a physical approximation. You do get slightly resolution but the color suffers greatly.

(by 1 chipper I am NOT referring to the new generation of single-chip sensors with three R-G-B layers, whch are essentially 3-chips in one.)

The statement quoted above that was made in my DVX thread about the DVX's CCD's being 934x768 was incorrect, and was originally made by me based on some information I found for Panasonic broadcast sensors...this was before i actually captured the entire raw frame.

Hope this helps,
Juan

[Rob Lohman]

Juan: do you happen to know where the limitation on this sample
rate is? Can the CCD block/array just not sample faster at higher
resolution or frame rates or is the problem in clocks or chips? Or
just a firmware limitation, perhaps?

If the CCD block just cannot read the information faster than 5
or 10 fps (in a digital still camera for example) then it doesn't
matter what we do to it, ofcourse.

Any thoughts on this?

[Juan P. Pertierra]

Rob,

There is a limitation on every CCD/CMOS on how fast it can read. This is the same concept why a line-scan camera with a line CCD spinning on a drum doesn't work, it just can't read that fast(and is not that sensitive).

Specially if you're talking about a Sony camera, you can find service manuals online easily and you can also find the data sheets for the CCD sensors. All sensors have a maximum rate at which they provide information, and most of them have different modes...for example, some of them do 10fps in full frame but can do 30fps in quarter frame, etc.

If the CCD cannot handle the sample rate you want, there is nothing you can do, like you said.

Juan