Why uprez the native CCD size?

[JerseyMike]

This isn't really an HVX-specific question, but it has to do with CCD resolution which seems to be a hot issue now.

Let's say the number of pixels is less than the final image resolution, what's the advantage to interpolating extra pixels in the camera versus just recording the native resolution of the CCD and handling the interpolation in post? Aren't you just consuming more P2/hard drive space by doing that in-cam?

Is there something in the interpolation process that must be done in the camera?

I really don't understand the interpolation process so forgive me if this is a dumb question.

[Anders Holck]

there several reasons to this:

1. foremost we must deliver a standard framesize to the codec, in our case DVCPRO HD. it can be either 960x720 or 1280x1080.

2. We can use vertical off-set aka. pixelshift. Interpolation is the creation of pixels that dont exist at all. By using pixelsshift, we actually have some information in some of the color channels to increase the size. But we cant use the pixelshift after compression as we would need the raw RGB data from the CCD.

[JerseyMike]

Ahh that makes perfect sense. I didn't even consider the problems with trying to interpolate after compression. Thanks Anders!

[BobDiaz]

Assume that the native resolution for the CCDs is 960h x 540v. These numbers are TOO LOW, but it make my example easy for follow. (There is evidence to that suggests that the native CCDs are higher in resolution than this.)

The Blue and Red CCDs are offset from the Green CCD by 1/2 a pixel horizontal and 1/2 a pixel vertical. If we look at the sum total of all the information from the CCDs, there are 1920 points of information in the horizontal and 1080 points of information in the vertical. Recording in native mode (960x540) would loose this additional information from the CCD offset.

OK, one could create a native mode recording that is NATIVE 4:4:4, but this generates a non-standard output that editing software must deal with.


Bob Diaz

[JerseyMike]

Hi Bob,
Thanks for the helpful information. Which leads me to my next question- Are any video camera makers planning to use Foveon's sensor anytime soon? I'm surprised it hasn't taken off yet.
Mike

[Anders Holck]

The Foveon 4.5 MP sensor will only do 640x480 at 30fps and 7 fps for 1440x1060.
Also the DSP is power to process the output is pretty steep and the Blue layer is pretty noisy and lacks sensitivity. Also there is a slight problem with skin tones because of the spectral response of the layers.

[BobDiaz]

In addition to the frame rate limition, I recall there was some problems with strong over lit colors. If a red car was too bright, the red would wash out all the other colors for the pixel(s) and create a white pixel(s). Very odd looking. Another odd issue was that lights in a night image has a weird blue glow around them. As for colors, the colors were OK, but a little off.

Foveon has a great idea, but it will be several generations before we see it working in video cameras.

Bob Diaz

[Anders Holck]

... I recall there was some problems with strong over lit colors. If a red car was too bright, the red would wash out all the other colors for the pixel(s) and create a white pixel(s). Very odd looking. Another odd issue was that lights in a night image has a weird blue glow around them.

Yes this was with the first generation of RAW converter software for the sensors, and seen by owners of the Sigma SD9 and SD10 SLR's. It was fixed in the later generation of converters, so I don't think it was a problem with the sensor but an insufficient algorithm instead.

The problem with the color response is due to the way the layers are filtering the light.
In a bayer chip or a prism based camera you can tune the exact filtering property of the primary colors. In th Foveon yuou have to compromise so the other frequences will pass thru to the other layers.

Dave Coffin, the programmmer of the dcraw RAW converter states:
"The Foveon X3 Capture chip requires a different kind of interpolation. Unlike CCD arrays, it captures three colors at every pixel location. But the colors are not well separated, so the raw data looks very gray. Much processing is needed to enhance color while suppressing noise."

It is seen as purple or green color cast, making skin tones look funny.