![]() There are several math errors in astropixel's post (multiplication and failure to take square root): More importantly it makes dark frame subtraction rather difficult if not impossible. Mike's tool mentioned above certainly didn't produce expected results for me due to these issues. So you need to be wary of using tools and techniques designed for unadulterated CDD data on very much adulterated DSLR data. This was discovered by Craig Stark and documented a few years ago, and I ran my own tests on my 500D, written up here: If you plot brightness against time or temperature it goes down, which is the opposite of what should happen. ![]() If you take dark frames and plot noise against increasing exposure time, or against fixed exposure time and increasing temperature (same difference), the noise increases as you'd expect. The accumulated dark current is measured from the optical black area and subtracted somehow (specific details of how are not available outside Canon as far as I know). You need to be aware that most Canon DSLRs (certainly of the 450D era) suppress dark current by pre-processing on camera before writing out the RAW file. Total noise ? dark frame -5?C 1800sec : -5?C 1800sec dark: 12.95*0.664 = 8.599 e-Įstimate the dark noise from the total noise of the -5?C bias and dark framesĮstimate dark noise standard deviation : -5?Cdarknoise = Sqrt(8.599^2 - 7.284^2) = 4.57 e-Įstimate dark signal due to dark current : -5?Cdarksignal = -5?Cdarknoise^2 = 20.886 e-Įstimate dark current : -5?Cdarkcurrent = -5?Cdarksignal / 1800 = 0.116 e-/sec Run the same exercise for -5?C bias and dark frames ![]() Total noise ? dark frame 20?C 1800sec - 48.07*0.664 = 31.918 e-Įstimate the dark noise from the total noise of the 20?C bias and dark framesĮstimate dark noise standard deviation : 20?CDarkNoise = Sqrt(31.918^2 ? 7.385^2) = 964.251 e-Įstimate dark signal due to dark current : 20?CDarkSignal = 20?CDarkNoise^2 = 929780.262 e-Įstimate dark current - 20?CDarkCurrent : 20?CDarkSignal/1800 = 516.545 e-/sec Over or under exposure will make estimates unreliable.Ĭalculate dark noise, dark signal and dark current Be sure your flats are well exposed, say to about 50% full well. PS: Based on other data I have seen, I think your gain estimate for ISO400 is reasonable, to within roughly +/- 10%. IMO this estimate is also suspect due to the relatively short dark exposure. I was not clear in my dark noise formula, do this instead:Ģ0CDarkNoise = Sqrt(6.84^2 - 6.72^2) = 1.27 e- (estimate of standard deviation of noise due to dark current)Ģ0CDarkSignal = 20CDarkNoise^2 = 1.63 e- (estimate of signal due to dark current)Ģ0CDarkCurrent = 20CDarkSignal / 180 = 0.009 e-/sec (dark current signal divided by time) But even with longer exposures, if offset drifts enough, this offset formula is not going to work at all. For low dark current detectors, it takes a while for a dark electron or two or more to appear in a pixel, so to be measured reliably. Dark exposures should be longer for reliable estimation, maybe 10 to 20 times longer. But offset does vary frame to frame due to detector electronics drift, and probably in this case the drift overwhelmed the increase due to dark current. This is typically the case as dark current increases offset. The offset forumla requires dark offset to be larger than bias offset. So detector 14-bit DN's are identical to PI's 16-bit DN's. This follows naturally from how PI reads and normalizes 14-bit data into the lower quarter of the normalized 0 to 1 range. Bit size is accounted for correctly by default in PI.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |