CCD Overscan Region.

[Tandum]

I'm trying to come to grips with this CCD feature and want someone that knows, to verify what I think :)

I understand this to be an area of storage within the sensor which is not connected to any pixels. Once the read of the sensor is done, the overscan area holds info on the read signal introduced by reading the sensor. Taking a mean of the overscan area allows you to remove this introduced signal from the image.

This differs to a bias signal which has a patern associated with it and is related to individual pixels. The Overscan level would therefore be classed as read noise and is uniformally removed from each pixel.

The overscan area should not respond to light.

Does that sound pretty close?

[marki]

Thats how I read it Robin . Do you think adjusting for it is worth the effort considering the low noise read?


Mark

[mjc]

I was going to post a question on this but thought it wise to see if there was previous interest - so tagging my comments here.

Robin
I was going to ask the question "does anyone on this forum use overscan?" - Because I'd like to get hold of a couple of frames to play with.

Did you get to grips with it?

In my understanding your comments are correct.

To refine a little more - and from what I read (I don't have hand-on experience) - there is, apparently, a uniform bias level across the chip - but this can vary across different exposures. There is also a pixel-to-pixel variation which is a said to be a constant offset per pixel to this uniform base-level. That is, the pattern is fixed but the overall values change in a uniform manner across the chip.

I'm guessing that the fixed pixel-pixel variations is due to fixed physical variations in the fabrication at pixel-level leading to different - but similar - resistances leading to fixed differences in voltages despite a uniform voltage across the chip. But I stress I'm guessing at that one.

The advantage of using the overscan region is that we measure the actual variable overscan from the overscan region of our bias frames and determine the constant pixel-pixel deviations by subtracting the overscan from the contents of the rest of the frame.

When we have our real (light) image we can subtract the uniform overscan (which we can extract from the overscan region from that frame) and then subtract the fixed pixel offsets from our previous bias image (which wass first overscan corrected).

Also once any frame has been ovsercan corrected it should be trimmed so that the overscan region(s) are cropped out of the image before any further processing.

In my understanding this is superior to just subtracting averaged bias frames as the latter is always going to just approximate what our bias level is in the light images.

I don't have the ability to extract overscan regions but if some kind soul could provide me with a handfull of bias-frames with overscan region still attached I'd like to get a little hands-on experience playing with them.

One thing which I'm curious about is that I suspect that the provision of overscan regions are more common than we might think - the chip on my Atik-16IC has some "black rows" and "black columns" (from data sheet) at the periphery of the chip so I suspect that it is only firmware and/or drivers that prevents the downloading of the overscan regions in consumer-grade CCD's.

I also note that SBIG appear to have recently (this year) provided drivers, in response to user feedback, to make the overscan region accessible on some/all of their imagers:

http://www.sbig.com/pdffiles/EnablingOverscanRegions.pdf

Does any of this interest or provoke further thought?


Mark

[Tandum]

Mark, the only other thing I found out is that on the 8300 sensor, which is what I was interested in, the overscan area is made up of real pixels on the sensor which are physically shielded from exposure to light by a metal shield. It is possible for proper imaging pixels to overflow into the overscan pixels rendering them useless, most often seen when combining pixel in a higher binning mode, and is due to the small well depth of the 8300 chips.