|
Licked - I think
OK - I think I understand my problem.
Its very simple really - and has nothing to do with flats, filters, LP etc etc.
My camera, like all CCDs, does not respond equally at all wavelengths of visible light. The camera is most sensitive in the green and tails off gently towards the red. In blue however, it has a rather sharper cut-off.
So when I take images of the same integration time, I have more signal in green, slightly less in red and much less in blue. How much depends on the colour of the target.
All image processing programs cater for this. Maxim allows manual adjustment to the channel weights when combining - the G2V star calibration provides a good guide to the appropriate weightings. PI allows for this too although the autostretch function in STF does this for you by allocating a weighting to R, G and B.
The key however is that the level of signal in B is still quite low and needs to be stretched to match the G and R ie normalised. This level of stretching is excessive, particularly in low signal areas - I guess where SNR is high. When this happens in the bright areas of the image - no problem as there is still plenty of signal to play with. But in the dark areas, there is less margin for error. The noise in the image gets stretched as well ie its brightness is boosted - so that areas that should be dark are in fact lighter than they need to be. The same thing happens in R and G but because the B is that much fainter, the stretch is that much greater. So I am boosting the noise much more in B than R and G.
This explains why the excess blue is in the fainter areas of the image - which happen to be near the edges of my images.
I am sure I'm not explaining this well. Hopefully the attached images illustrate the issue.
The images are, in order, raw uncalibrated frames in R, G and B respectively. Next to them is the stats for the frame. Note how much lower the Median is for the B frame ie overall it is much darker. I have stretched each image by the same amount. The brightness of the R and G are similar while the B is much dimmer. There is less dynamic range to work with so the dark areas get dragged up too much - resulting in a blue hue to the shadows.
Now I am sure that applying a flat to the images complicates things. Unlike the lights, the integration times for the flats are scaled. This is because the integration time is matched across all filters by the ADU in each flat. So that I have normalised the flats through changing the integration times. In contrast, the lights are not normalised in this way. So - I think- that when PI flat calibrates, it applies a high dynamic range flat to a low dynamic range light. So flat calibration compounds the problem.
Does this all make sense?
I picked this up while playing with the raw subs. At the suggestion of a wise mate I combined a single R, G and B frame of 180 seconds - uncalibrated - in a variety of orientations (rotating filters, rotating the camera etc) and found no real pattern other than that the excess blue was always in the shadows.
So how to deal with this? I have 2 ideas.
The first is to simply up the integration times for the B filter. Taking a few G2V measurements, my weightings should be roughly 1:1:1.6. So I increase my B subs to 288 seconds from 180 seconds. That should give me a better dynamic range (and reduce the SNR in the darker areas). I can then let PI autostretch and use DBE to clean up any residual gradient.
The second is to increase the integration times across all subs. This seems less attractive for me. I am near the Sydney CBD and LP is heavy duty here. I'm also near a flight path and longer subs means greater prospects of plane trails.
I'd appreciate any thoughts on this - but I'm pretty sure I'm on the right track. Among other things, it explains why the colour balance in bright areas of my images are usually OK.
Pete
|