None. E.g. Canon D300 mentioned by me has dynamic range not lower than film while in RAW mode, and I guess even better.

what?? are you sure about this? last time I looked film had like 500% more dynamic range then video...I am sure my canon 10D has less then film and it's a professional SLR 6mp camera.....

I think there are two different things we might be thinking about...one is the smallest difference in color(luminosity for one CCD/CMOS) that can be represented...in scott's book Film is shown as 16-bit RGB equivalent, so if you had 16-bit A/D's, technically you should have the same precision in luminosity,

however, another difference is what maximum level of brightness can be shown which i think is somewhat referred to as latitude(?), i.e. at which point film shows 'all white' versus a CCD. In this case, i think all CCD's and CMOS so far clip at a much lower brightness than film...i think this is in part due because film has a logarithmic response to light intensity while CCD's/CMOS are linear.

I have heard over and over, though, that the more precision in a digital aquisition system(i.e. wider a/d's) you can get a better result when applying a logarithmic(or 'S' curve) because you can pull more detail from the darks which would otherwise not be there on a lower precision system.

Once again, i'm not an expert on this, this is my best understanding of it so far....

Juan

that sounds right to me Juan...it is the brightness I was talking about.... I want to see some frame grabs at 30bit!

ok, i am almost 100% this is a 30-bit capture. Note that there are still 6 bits missing because of the limitations of my capture card, so this image has 83% of the total information.

Recall that this is 0.9 pixel aspect ratio, so it looks a bit stretched unless your viewer compensates for it.

There is a hefty amount of noise, which is NOT due to the standard thermal noise in the CCD's, but it is due to my test setup. it's less than before but still there.

I tried to make a large range of illumination, and all i can say is that on the DV footage I cannot see anything in the dark area. I can't see any thermal noise, but that is probably due to the lower 6 bits missing.

Also, this is my 'best guess' in color balance. This is probably completely off, i know nothing about colorimetry. Maybe someone can offer to put the three R,G,B frames together correctly?

http://expert.cc.purdue.edu/~pertierr/dvxcap1.tif

Juan.

For best comparison, please attach a 8-bit normal DV frame to EACH 10 or 12-bit frame you present. That would let us see the difeference. Just simply grab the DV footage and transform it to BMP with all the limitations and arfifacts it has.

PS. The 16-bit TIFF frame that you showed us is not correctly lit. Just pay attention to the histogram of it. most of the information is has in it's lower tonal parts. Please be more correct adjusting exposure of your dvx.

well, this is why i wanted someone who knows what they are doing to put together the image from the RGB frames...also, i think i mentioned it before, i do not have control over exposure yet, because the camera is open and that control is disconnected, so all i have is the 100W bulb i was using for lighting, and i made it as bright as possible without clipping.

Juan

Juan.

Just pay attention to the lighting. Can you properly read the historgram? Check it. So far the image is clearly underexposed. If you have just one light, try to take it closer to the subject so to get more lighting. And don't be afraid of overexposion as this could tell us much when comparing the RAW frame to DV frame.

And grab the DV frame for comparison.

The best test comparison frame would be preparing the scene that one part of it would be overexposed in terms of tipical DV footage; the second part of it would be quite good exposed and the third part of it - underexposed. Prepare it:
1. as a DV frame
2. as a RAW frame.

Don't care about balance as it is a second rate thing. We could correct it in photoshop for our own.

Ok, here's an attempt at what you ask.

These are exactly the same frame, captured through DV and captured RAW.

Now, along with the exposure scroller, the WB button is also disconnected so that is why it all looks so red, but you can see the difference anyway. The menu button is also disconnected so i couldn't take letterbox mode off.

I lit it from the side and had one section overexposed(in the DV viewfinder).

Note there's a lot of speckled noise in the RAW capture as usual, and this is 30bits of 36.

http://expert.cc.purdue.edu/~pertierr/dvxcaptest1_DV.tif

http://expert.cc.purdue.edu/~pertierr/dvxcaptest1_RAW.tif

Juan, thank you. That made it.

So it looks as if indeed the tonal dynamics of the RAW exeeds the DV dynamics and as hell it is a good news, making your work Juan not only a hobbyst play but a serious serious thing to turn a prosumer camera into a real high end thing.

I notice there is a lot more additional tonal range in the overexposed area than in underexposed area. Actually, the shadow information in RAW and in DV is simillar. I think this is easy to explain and is not a problem in further shooting using your equipment. The engeneers working on the camera had to make a decision of what part of this wide 12-bit tonal range put into the 8-bit DV output. What would be the choose? Of course the lower part as it makes the whole camera most sensitive.

I guess while shooting in 12-bit it would be best to overexpose the image by 1 or 2 f-stops to gain some more shadow details while still keeping some additional details in light areas over the DV. After all everything could be corrected in post production.

AND THAT IS GREAT.

Juan - tip: When you light the scene, the viewfinder of the camera as well as zebra and all the camera measure tools would be misleading. Actually to properly expose the 12-bit image you should overexpose (in terms of DV) your scene - otherwise we will not see any advantages of the 12 bit image over DV in terms of dynamic range. Just set the lighting with the HISTOGRAM!!! It is clearly seen on the histogram that the RAW image that you sent is still underexposed!!! MORE LIGHT!!!!

Of course, this will damage the DV footage but to hell with it.

OK I just mastered the RAW frame that you sent.

The noise. The thing I found out is that it is NOT random, some distingued areas are just clearly covered by it and others not. Actually I guess this has something to do with spatial frequrency. Those dots are more in areas where color changes most. So, is it really problem with capture card or something deeper? Would capture card react in such a way to high frequency changing not of data rate but information itself??? And pay attention that the noise is made of a pixels wich somewhat lacks R or G component. I never found a noise pixel that lacks B component!!! Maybe that could be a hint?

Milosz,

I have noticed the pattern too, so maybe it is still an intermittent connection? It seems like the speckles happen when a specific bit just goes to zero, and it seems to be only higher-end bits such that when they go to zero, it sets the color very dark...also the fact that it is a specific bit goes along with the fact that it only occurs in areas of similar color, or areas where the color changes through those values.

Maybe I can still fix this problem....if i could only find an exact pattern.

Juan

My suggestion is find out why this happens only to R and G components while the B component is always ok.

Actually you don't need to overexpose the camera by two stops, that's the whole point of using a higher bit-depth image. There's a TON of shadow detail to lift up if you apply the correct curve to the image. Using curves, you can get a really nice image that keeps both the shadows and the highlights. Also if I'm not mistaken, there's probably another stop in the shadows if he's missing the lower two bits of the image. Either way, I think you should be using the zebras, as they will give you an exposure that will capture all the shadow detail without burning out the highlights-maybe you could overexpose by half-a stop over 100 zebra, but I wouldn't go much futher than that, because you don't know when the highlights in the RAW files will burn out, and when they do that's bad. And you don't have to worry about banding with the 12-bit images when you "lift" up the shadows. So I don't really see the harm in underexposing by a bit.

http://www.dv.com/jive3/thread.jspa?threadID=300001061&tstart=15

http://home.teleport.com/~gdi/vancecam.htm

Here’s a guy who, like many of us, thought the off the shelf offerings just wouldn’t hack it.

His approach is a bit different but he has a lot of good ideas. His approach is to treat the camcorder for what it is a cobbled together collection of technology, a systems approach.

When he finished he had something that was actually useable.


Just a bit off topic because it's not a DVX100 based system but its cool none the less.

There's probably going to be a remarkable amount of color precision in those missing 6-bits. Right now there are 1024 levels per channel. When the lower 2-bits/channel are available there will be 4096 levels...technically there should be (4096 * 3) colors available which is a lot of color, but i think RGB has a lot of overlapping information such that the actual number is less.

However, i'm not sure if there will be that much more latitude in the lower bits, just because they are low-end bits. For example, right now adding the two bits will only add a value of 3 to the maximum number of levels that can be represented per channel...however these lower bits add 4 intermediate levels between every two levels available right now in the 10-bit RGB, which is a very considerable amount of additional precision.

I never expected the raw output to have more latitude than the DV output, but I stand corrected! I guess the people at Panasonic had to do some incredible compromises just to make it look half decent on DV.

It's like buying a ferrari with the hand brake permanently set.

Jason:

I would wonder if you are right. By >>overexposing<< I was refering to the DV footage, but in terms of RAW image - opening some f-stops would turn the image from >>underexposed<< to >>properly exposed<<.

There is some thing that tells me that you don't understand the thing at all: you are talking of exposing to zebra - but the zebra's standard DV 100% of exposure is actually at the level of about 30 percent tonal in the RAW image. Grabbing RAW, and adjusting exposure to this standard DV 100% of zebra would be wasting 70% of RAW capacity. Actually if you talk about zebra, we would rather need a sort of "RAW ZEBRA", that could tell us when the whites are clipping not in DV range, but in RAW range that we work on.

Of course, I'm kidding with this "RAW ZEBRA". It would be just enough to - firstly - set the exposure to the standard DV zebra 100% and then lift the exposure by a factor of 1 or 2 f-stops, how much this should be - that would tests of Mr Juan show.

And why to expose with more light?

1. More information in shadows (while not clipping whites!!! - so no loss of information in highlites)
2. LESS NOISE IN SHADOWS.

Of course there is always a question if those upper 70% space of exposure in RAW is sort of the same quality as the lower 30% in which the camera exposes normally to DV. It could turn out that this upper range is somewhat inferior to the "normal", I don't know - not linear or something, but right now no one of us knows this until Juan makes some test (preferably with a person's face, or maybe showing a sunny street would be a good idea anyway). If it is not inferior, than I'm sure on my point.


Juan:

And this is frustrating that buying better model you overpay just for the brake to be turned off:)

When I follow Juan's links to the DV and Raw files, I can get them in my internet browser, but I can't seem to find a way to pull them out into any other program to play with them. It loads in Quicktime, and when I click on the image, it just gives me quicktime options.
Is there a trick to getting them to check them out?

I feel like there is probably a really easy answer to this, and I am just overlooking it. Thanks for the help.

Try to download the file with a download application as GetRight or Dowload Accelerator.

I've activated the links so you can right-click on them and then
download to your harddisk for opening in other applications.

Juan: to do this yourself in the future take a look at this page (http://www.dvinfo.net/conf/misc.php?s=&action=bbcode)
in our forum guide. It explains how to add url tags around your
links so they get hyperlinked.

Thanks Rob! I meant to do it but totally forgot, will do it next time.

I'm in the process of moving the camera right now...

I managed to move the entire thing, computer and all to my sliding door window, but i got done pretty late last night...it's a nice day out today so i should have a clip soon.

I'm also taking a stab at reseating the probe to see if the speckles are related to the connections. As far as i can tell there are two bits one in the green an another in the red which somewhat randomly goes to zero and sets the pixels to a dark color...

if someone can take a stab at looking at the speckles and seeing if they agree, and which bits are causing the problem, i'd appreciate it just to have some other insight. I checked them by looking at adjacent pixel values around the speckles and then seeing what was missing from the speckled value...think about it in binary numbers, and if a single bit is causing the problem all over an area....

Thanks
Juan

Ok, i have a couple of clips. Not much just a car driving by my house and a friend of mine waving. I also have DV counterparts. Like the previous frames, the difference in latitude is obvious. In one shot, the sky, concrete road, and several other areas of the scene are just straight out white on the DV footage, while nothing clips in the RAW footage, and even captures the different tones in the sky.

However, a 3 second clip is well over 100MB, and I do not have the web space. What does everyone suggest? Also, i have a stream of frames which i can use as an animation in Shake, or FCP but does anyone have any suggestions as to what video format to use? The best I can do with what I have is 4:2:2 uncompressed 10-bit MOV i think.

I've got all 30-bits, but the speckled noise is still there...my main task now is to try and figure out where it is coming from.

Juan

I've got space but no bandwith left because of my test clips. I know DVInfo does or I could take donations for extra bandwith if they don't have space.

I may have about 4~5 gb bandwidth a day to spare for a while. Let me know if you need it and I can setup an ftp account for you.

-Rodger

I have posted 30-bit RAW and DV versions of the same frame, from a clip i recorded today. Unfortunately, the speckles affected nearly all the fine details in the RAW capture, but nevertheless it is a good example of the added latitude.

The color is waaaaay of(can you guess from the red sky?), but this is because I am posting the unaltered output from the CCD's...if you want to take a crack at doing color correction please do so and email me the result so i can post it.

http://expert.cc.purdue.edu/~pertierr/cap2_DV.tif

http://expert.cc.purdue.edu/~pertierr/cap2_RAW.tif

I am hard at work now trying to figure out where this video-pox is coming from. :)

Juan,

The R,G,B channels in the RAW data don't seem to be lining up.

In Shake view the separate channels of both the DV and the RAW and you can see the shifting in the RAW.

That would explain the colored edge fringing.

Shooting a b&w chart would help figure out the pixel shifting.

Great work. I monitor this thread very closely.

Pete

That's right! Thanks for pointing it out.

It's tricky to find what is the correct alignment, so i decided to post the photoshop file for that frame, with the R,G,B as separate layers. If any of you can take a look and let me know what you think is the correct alignment. I beleive the infamous pink line on the left of the DVX output may be just an artifact of shifting the R,G,B frames for alignment, so it might be a hint as to what is correct. This psd file works well also if any of you want to try to color correct this thing:

http://expert.cc.purdue.edu/~pertierr/cap2_RAW.psd

I'm assuming the color issues is due to the camera not doing the
whitebalancing to after you captured the RAW footage? If so,
some on the fly color shifting (something else than correcting)
might be interesting to have...

I did a little manual color balancing in Vegas (somehow it doesn't
seem to have a whitebalance filter and neither does Paintshop
Pro, weird!) and came up with the following:

manual white balanced cap2_RAW (www.visuar.com/DVi/cap2_WB.tif)

Here's a first pass at lining up the channels. Using the green channel as the base to match.

red x-shift = -0.7 y-shift = 0.7
grn x-shift = 0.0 y-shift = 0.0
blu x-shift = -1.0 y-shift = -0.5

Unfortunately it's sub-pixel movement. Are you shifting the channels around in your software?

Nope, i'm posting the frames just as i receive them from the A/D's, and just aligning them with 0 offset.

I was afraid that in fact the CCD alignment on the prism is not perfect and fractional pixel offsets would be involved....maybe if we up-res it to HD, and ~then~ overlay them and shift them?

Juan

I also did little manual channels adjusting, color correction, curves and alignining in Photoshop.

Here's the result:

http://www.starzonefinland.net/jarno/cap2_RAW_CC_ALIGN.tif

That's nearly as good as I could get it, unfortunately you still can see some noise in the contrasts. It _looks_ like the channels are still unaligned, but while trying to correct it I found out it's sub-pixel, just like you wrote and mixed with artificial noise.

Oh well, hope this helps. I can see if there is still something i could do with the frame, but for now i'll have to leave it alone and go read some internal medicine... And yeah, you're doing a great job - keep on up with it.

-Jarno

ok, i need some help here.

I'm trying to get rid of the noise, and in examining the raw data files along with the images i'm getting, there is something i don't understand.

I am using only the high 10 bits of 16-bits, there for the highest number that can be represented is 1111111111000000b, and the lowest is 0b.

However when i load the image as a raw file in photoshop or image converter, it reads a 68 decimal value as a grayscale level of ~4. Do paint programs set the black value at 64decimal or so? Like wise, what is considered 'white'? because even though I can reach the full 16-bit range because i'm only using 10-bits, if i overexpose something i still get white in photoshop!

I'm burned out for tonight trying to figure out where these speckles are coming from...please do give this a crack:

if i aim the camera at a light and overexpose the entire frame, i get ALL WHITE, no speckles! What kind of 'noise' is so specific? The whole theory of a high-order bit being intermittent doesn't work because for white all bits have to be active, if any bits where intermittent or badly connected it would surely show up as noise.

It's a brain-frier. I any of you have any theories whatsoever i'd like to hear them.

Juan

Here is the most speckle-free 4:4:4 30-bit frame i've gotten out of it so far. This one was a very dark frame to begin with(ND2), i adjusted the levels a little. I like the way the dark details still showed so i here it is:

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

I ~think~ the noise problem might be a ground problem, but i'm not sure. I did some tests which seemed to reduce the noise but the sun was also setting, so I am not sure if the problem is just with the mid-range luminosity.

Juan the spots are what looks sorta like what is called a dead pixel ...I am guessing it is caused by some sort of bad connection you have on the test bench

Obin,

I doubt it is a dead pixel, or even a hot pixel issue.

Juan pointed out that the noise seemed to go away when he overexposed. If the pixels were dead (no charge) or hot (always charged) they would still show up as noise when over or under exposed.

I don't know enough to know what the problem may be...but I can at least guess at what it isn't
:)

-Luis

I did again some testing with the cap5 frame, and here's what I got:

http://www.starzonefinland.net/jarno/cap5_BMP.bmp

It's just a 24-bit bitmap, but what's essential are the color samples up there; they represent with a ~60 percent accuracy the noise hues, that is, if you in Photoshop select the color samples with a 60% magic wand, you get almost all of the visible noise selected.

What seems to be quite interesting is that the colors seem to line up somehow to represent some kind of a color scale. I didn't have time to do any calculations, but I think they're somewhat connected...

Well, I'll return with this issue later.

-Jarno

PS, does your browser show the tiff images correctly? My computer opens them with a Quicktime extension, and they look false, for example the cap5.tiff seemed okay in PS but in explorer it was all blue? Is it just that QT doesn't support the 16-bit images?

Jarno:

That's right, QT, does not support 16-bit and shows them false in the browser.

Jarno,

That is an interesting point, thanks for noting that...any other relationship you can come up with will greatly help me in figuring this out.

Another interesting point: I am synching up my capture card on the PC with the clock that drives the three A/D's. The images i've shown you so far are capturing using the rising edge of the clock. if I switch to sync with the falling edge of the clock, i get the same images but with ~more~ speckles in the same areas as before. This makes me think that it has something to do with the clock, maybe i am driving too much stuff with it and it is degrading.

Juan

I done calculated the noise colors.... Arrgh. This makes me nuts. I couldn't find any greater correlation between the color values, at least with my math. My friend is studying maths, maybe she could figure something out of the values. That shall be seen tomorrow, now I'll anyway go to sleep, it's 2.00am and my head's spinning around with the values... :)

-Jarno

btw, found out a funny thing. If you convert the picture (cap2_RAW) to lab color and turn off the 'a'-channel, you get a quite nice looking image. Also in CMYK, if you turn off the magenta channel, the image looks quite correct (a bit greenish), without having to do any other color balancing. Anyway, probably this hasn't anything to do with nothing, just a detail...

Wow. This stuff is just...hot DANG, this is cool!

Now, pardon my ignorant question (I'm no electrical engineer), but these speckles we see in the example image...could this be because of what one might consider "broken" CCDs?

I get the image in my head of JVC saying "We've got thousands of these chips sitting around that didn't pass the quality tests for our higher end cameras, why not throw 'em in the DV line?" Seems to me all they'd have to do is interpolate the values of the noisy pixels--selected using those color sample blobs--from the ones around them. I was able to eliminate all the noticeable noise using the simple Color Replacer in Paint Shop Pro, so it must not be THAT hard, right?

The image IS going to be compressed, after all, and recorded at a lower resolution, so imperfections in the initial image might not make it through all the way...

Is this a realistic scenario?

I'm pretty sure it is not broken CCD's because the speckles change locations in every frame, although they are located in the same general area, besides whether i sample on the rising or falling edge of the clock seems to affect the quantity of speckles...so it seems to be clock related.

Juan

Gotcha.

Can't wait to see where this all leads...very exciting concept. Best of luck with your experimentation!

Juan,

Is it possible that Robert is on to something here?

I'm not saying it is bad CCD chips necessarily...

But, you aren't seeing any of this noise on the DV tape are you?
Is it all being filtered out in some way through the DV compression?

Is is possible that the noise is always there, on all DVX100s?
But, as Robert mentioned, it wouldn't matter because it's getting thrown out anyhow?

Probably not...but I figured it was worth mentioning.
You haven't seen any noise on the DV frames, right?

-Luis

I don't think that the noise is supposed to be there. Despite DV's compression ratio, it doesn't turn out a horrible picture, so a dark blue pixel will still be a dark blue pixel. And since there aren't that many extra vertical lines of resolution (what's the full res? 771x492 or something?), it can't all be interpolated away. And finally, the compression itself wouldn't cover for it because such a sharp change in contrast would be expected to be preserved by a compression scheme, not eliminated (imagine what it would do to text otherwise).

My guess is that he's got one of the higher bits loose on the CCDs' connections. If the bit is dead or intermittent then when everything lines up just right and that bit dies, the color value will drop to very low. For instance: if the color of a pixel of the sky is 1000000111 (519 decimal) and bit 10 (assuming big-endian) is flaky and happens to not make the connection on that clock cycle, then the color of the pixel would be returned as 0000000111 (7 decimal) which would return a very very dark pixel compared to what was supposed to be there.

And this makes perfect sense, since from what I have seen from the pictures, it's the very middle of the color range that's having the majority of the noise problem (and 512 is right in the middle of the color space) where you drop one bit and the color is changed dramatically.


Although, to be sure, the best thing to do would be to get another camera and try it on there as well to see if it's Juan's CCDs or if it's his method.

Jon just pointed out what has been my best theory so far...however, i am pretty sure at this point that all the connections are solid and not physically intermittent....so my best guess at this point is that the clock signal which already drives the three A/D's, is degrading because I am also driving the capture card with it...i'm going to try and make a clock-follower that syncs up to the original clock and see if that helps.

Juan

Well, I can see I'm out of my league with the noise theory, but I've got a couple other questions:

1.) Assuming that, once all is said and done, this procedure were possible on other cameras (let's say, the VX2000), how difficult is installation for a layman such as myself? Yes, yes, I know, you could tell me "You've never done anything worthwhile with your STANDARD camera, let alone with one of these fancy mods--focus on your content before you worry about your image", and you'd be right.

But just out of curiosity. Will one need a degree to do this? I mean, I've disassembled my camera several times, and have become familiar with the location of boards, screws, and the like, along with the required order of dis/reassembly, but I couldn't tell ya what chips perform which functions.

2.) What's the legal status of this project? Don't get me wrong, I see no MORAL problem here. I think this is a great idea, and am anxious to see some final results, but I worry about the ramifications of something like this should the companies catch on.

Sorta like the Linux on XBox project. I use this example 'cause I'm a videogame nerd, and I think it's analogous enough to work (keep in mind, however, that I can't remember if this scenario actually happened, or was mere speculation).

The idea was that the XBox, being essentially a mini computer, would make a perfect homegrown Linux platform, especially with its low price point. The fact that it's made by Microsoft was even more incentive for interested parties to get this done. Well, apparently someone figured it out, and in the process of telling others how to do this mod, got a cease-and-desist from MS. The explanation, as I remember it, was that while it was perfectly all right to modify the unit that you paid for, and do whatever you like to the thing, it was against the law to distribute this information to others, in spite of the fact that anyone wanting to do this still had to purchase their product to perform the modification.

Whether this was a rumor, I cannot say for sure, but I think it's worth mentioning, as you seem to be treading similar waters here.

Will this be an issue for your product?

Don't you basically just void your warranty when you open the unit up?

I'm not an expert on the legal ramifications, but this process doesn't seem that different than the BBC modification to the sound units on VX2000 and PD150's. The BBC hasn't been sued for that either. Perhaps because they're a corporation and not an individual. I don't know.

I say worry about it later. Microsoft was probably more concerned that the Xbox was being outfitted with Linux instead of Windows XP. Sort of like setting up the mayor's daughter with a convict.