Ok - first of all a caveat... I did this because I was interested in the problem that Eric and Gabrieli reported. As a member of the research team here, I have some latitude to spend my time like this. And I hope you guys find it useful/interesting. However... please don't get me in trouble by asking me to do this on more Point Grey cameras, or by sending this forum article to our support department and say "please do what Don did for me for camera X". That'll probably get me in trouble
Ok? Please please?
Regarding those "bars"... sure enough, I reproduced the problem. It is always comforting to know that the world of engineering is repeatable, even when it produces results you wish were not there.
What I did was I set up a flat-field generator that I could illuminate with different single-spectrum LEDs. The device has an aperature of about F/8, with an exit pupil of about 10cm. I played around with grabbing images from an ICX674 camera and an ICX625 camera. I set the gain to 0 and turned off gamma. At first I couldn't see any problem. The images looked very flat to me. I tried with different LEDs but still couldn't see the issue. So I saved some images and tried magnifying the differences.
What I did was
- grabbed an 8-bit image (sorry, I was grabbing in 8 bit rather than 16) where the target image average was 200 greyscales
- smoothed these images substantially so I could isolate systematic changes in the image level v.s. read noise
- rescaled the images to be between 185 and 210 greyscales. That is, everything below 185 in the original image was scaled to 0, everything above 210 was scaled to 255, and everything between 185 to 210 was stretched to 0 to 255. This was to exaggerate differences in pixels
Lo and behold - I can see those bands.
I tried 3 light conditions - 652 nm (near H-a), 600-470 ("white light" - well, maybe a bit on the green/cyan side) and 416nm was as close as I had to CaK.
For the ICX674, the bands jump out in 652, they are somewhat reduced at 416, and they are all but gone in the "white light". For the ICX625 they are not there. There is some changes in signal across the sensor but it is nice and smooth. (Notice I cleaned the optical window on my ICX674 but didn't do that for the ICX625 - so please excuse the dust).
So then I went to talk to one of our sensor engineers and showed him my results and he said "yep, that is video shading, that is expected". (Note, before getting to this stage I considered a lot of other things like reflections in the cover glass, mechanical vignetting from the case, etc, without a lot of satisfaction. So it was nice to get a definitive comment from someone.)
He went on to describe to me how this occurs in the wafer fabrication process (done by Sony). Different layers are deposited across the face of the sensor with the goal that they are perfectly flat. However, near the edge of the sensor where the sensor "package" (what looks like black plastic running around the sensor active grid) there is often a change in height, and this makes it difficult to deposit a flat layer. Since there is a step in the base material, some layer material often "ramps up" slightly along this step. This causes a "frame shading" effect around the sensor active area. Different sensors will have a different height of the "step" depending on their design, so seeing differences between sensor models is normal.
Sony tests the sensors to ensure they pass a performance specification before selling these sensors. These sensors would all pass. I should point out that the framing variation of the ICX674 with WL is on the order of 0.4% - i.e. 1 greyscale. For 652 nm it is higher - about 4 or 5 greyscales at maximum and averaging about 2 or 3 along those bands. (Note in my smoothed and stretched images, each contour line you see is 1 greyscale of the sensor response).
I am sure Sony engineers would respond to feedback on this issue with a comment like "that is what flatfield compensation is for". Which doesn't really make your lives easier.
I'm not sure where you can go with this information. It is a sensor artifact based on the Sony chip and you will find it in any camera model that uses that chip (
This is the same image with the scaling to magnify the problem.
This is the ICX674 with "white" light, scaled.
ICX674 with 416 nm light
ICX625 with 652 nm light
ICX625 with "white" light