4:4:4 12-bit Uncompressed DVX100

[Peter Plevritis]

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?

[Juan P. Pertierra]

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

[Jarno Satopaa]

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

Here's the result:

http://www.starzonefinland.net/jarno...W_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

[Juan P. Pertierra]

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!

[Juan P. Pertierra]

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

[Juan P. Pertierra]

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.

[Obin Olson]

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

[Luis Caffesse]

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

[Jarno Satopaa]

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?

[Stephen van Vuuren]

Jarno:

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

[Juan P. Pertierra]

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

[Jarno Satopaa]

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...

[Robert Martens]

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?

[Juan P. Pertierra]

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

[Robert Martens]

Gotcha.

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