Canon 1000D EXIF temperature vs Cooling

[astropixel]

Perhaps this exercise would have been useful firsthand, investigating cooled DSLR calibration. Now that I understand the issues a little more, here is an exiftool dump of cooled 1000D @ -5C. This file indicates a camera temperature of 12C, whereas the CMOS is close to -5C, measured at the cold finger immediately adjacent to the CMOS.

Only one file here, but sampling several files, camera temperature increased to 15C over the session; ~17 - 20C difference in temperatures. During the 200 bias frame session camera temperature range was 13C to 18C - CMOS, near enough to -5C.

EDIT: Because the 1000D has a DIGIC II processor, does camera temperature == sensor temperature or a derived sensor temperature? Either way, and if camera temperature reported by EXIF is a function used to calculate firmware adjustments to RAW data, the disparity is significant. Sure, the dark current is less, but the correction applied to the frame may be excessive.

[IanL]

I haven't had time to get back to my own tests yet, but:

- I don't think the camera pre-processing uses the EXIF temperature at all. I suspect it uses the brightness of the optical black area to calibrate the adjustments. (Part of the sensor will be coated to block out light, and by measuring the brightness of this area the camera can determine the average dark current and make an adjustment.  This isn't useful to us as it is not a per-pixel measurement of the exposed area of the sensor.  I don't think the data from this area makes it in to the RAW file anyway, but others may know better?)

- One way to test this would be to plot the MADs (X axis) vs Medians (Y axis) of the frames on a graph. (As advised by Juan in my thread - I haven't done this yet for my own data).

- For your cooled camera I would expect you will see all the data clustered in one small area: Since the MADs should increase with increasing sensor temperature they should be close in value if the sensor is at a constant -5C; this means that the original medians would also be close in value. Assuming the camera processor is relying on the median of the optical black, any adjustment applied should also be similar across frames, thus leading to a similar median after processing.

- If you get a significant line of some sort on the graph, or a wide scattering of points, then that would indicate that the camera is using some other measurement to calculate the adjustment.

[Phil Leigh]

In-camera processing does not use the internal temperature sensor (which as you've discovered does not measure the CMOS temp anyway - that is not its purpose). It uses the data recorded in the sensor overscan (optical black) area.

No process - internal or external - should ever use the EXIF temp since it is not directly/linearly related to anything in the image data. Fortunately, PI doesn't use the EXIF temp for anything.