 |
Low-light integration and bright pixels
hibou from France [30 posts] |
17 year
|
Image integration under low-light conditions with uncooled CCD's produces a few very bright (noisy) pixels (example). But "Image Integration... will... normalize the image based on the lightest pixel seen in the image". It would be better to be able exclude isolated bright pixels (and isolated dark pixels) from this normalisation. I guess that this might be overcome by subtracting a dark field image, but it would still be nice to have this simple exclusion option.
Unrelated comment: it would be good to reset the integration to zero when the "Run" button is clicked off. The only way I found to reset was by changing the number of frames for the integration. It would also be nice if a flag came up when the integration had finished, or maybe have a running display of the number of frames accumulated.
|
|
|
|
Image?
Anonymous |
17 year
|
Alan,
Thanks! Great points! We're currently adding those in. Could you upload an image in which the isolated points of light cause the normalization to be off? We're not sure what those isolated pixels look like. I.e. how steep is the peak that they are on. It sounds like a non-maximal technique might do it but without an image to test we can't be sure.
Thanks,
STeven.
|
|
|
|
Low-light integration and bright pixels
hibou from France [30 posts] |
17 year
|
Thanks Steven. First let me show what I am trying to do. I have a camera that integrates electronically (Watec WAT-120N) and am imaging a scintillator screen. Here is an electronic integration of 256 frames with minimal integration in RoboRealm<br>
 <br>
Note the strong central spot with high contrast plus a few isolated hot pixels (this camera is not cooled). The question is, could I instead use a less expensive non-integrating camera and do all the integration in RoboRealm? OK, here is 1 frame from the camera with 300 frame integration in RoboRealm<br>
 <br>
Now the contrast is less high and the image much weaker. Integrating over more frames does not help. In fact more integration just smooths the image that I get with minimal 2-frame software integration<br>
 <br>
So then I tried to tweak the contrast, equalisation etc in RoboRealm. The following image is about as good as I can get<br>
 <br>
This is better, but clearly I am doing something wrong here, since it is still not as good as simple electronic integration within the camera.
My feeling is that RoboRealm's automatic normalisation to the brightest and darkest pixels might be the problem, since a few pixels are hot (and maybe a few are dead). In another image integrating application (AstroVideo used by amature astronomers for low-light imaging) you can set the zero and maximum levels for normalisation to exclude bright and dark pixels, and that did help considerably. I suppose that you could have input for these two additional parameters in the RoboRealm "integrate" and "average" dialogues.
Finally I played around with a cheap low-light WebCam (Philips SPC-900NC) which again has hot pixels for long high gain exposures. RoboRealm's integration did help a lot, but I felt that I should be able to do even better. Indeed I did manage to massage my low-light image using ACDSee, but off-line massaging defeats the purpose of having an optimum real-time image such as I can get with the integrating video camera.
Alan (hope I managed to integrate my images into this text at least :-)
|
|
|
|
Integrate Module Updated
Anonymous |
17 year
|
Alan,
Thanks for the images and explainations. That helped a lot. We've made a couple changes to that module and added most of the items you mentioned. We ran the image you got from RoboRealm (the second one above) and we got this result
with a 60, 40 ignore range. I think this may be closer to what you are looking for. Let us know ...
STeven.
|
|
|
|
Anonymous |
17 year
|
Oh, and please download RoboRealm again from the same url .. those modifications have already been uploaded to the website. :-)
Note that adding in a mean blur filter seems to soften the image a little so it looks more like the one you provided.
STeven.
|
|
|
|
Re: Integrate Module Updated
hibou from France [30 posts] |
17 year
|
Good work STeven. I downloaded the new version and tested it with my low-light Philips WebCam SPC-900NC. It now resets when you click Run off/on (great) and you have added to the Integrate module "Subtract Average" (extremely useful) and "Normalise Ignore Above/Below" controls (couldn't get this to work - see below).
1) Can we do the same for the Average module?
2) The raw test frame looks faint as expected when I drastically reduce light
3) When I integrate 20 frames and subtract the average but DON'T normalise I get a much better result (great so far). I could also smooth as you say.
4) If I integrate 100 frames and subtract the average, it all washes out. Clearly I need to normalise.
5) But if I normalise I only see the hot pixels. This is 100 frames 60/40 but even 100/0 gives much the same. And it makes little difference if I subtract the average, or change the number of frames etc
6) Curiously I get a negative image with 100 frames, subtract the average and normalise 0/90 (this combination should not even be possible).
So subtracting the average looks extremely promising but there seems to be some kind of strange bug in the normalisation. (BTW setting the number of frames to integrate only works if I use selection of the preset values, not if I type directly into the frames box - not a big problem).
I am excited about how how well this works and how easy it is to use despite my normalisation problem. Thanks again STeven. Alan.
|
|
|
|
Normalization Fixed
Anonymous |
17 year
|
There was a problem in the normalization routine. This has been fixed.
We also added the reset button and frames averaged to the average module.
The subtract average from the integration will help to reduce the noise in the image but will not prevent the need for normalization. Now that normalization is fixed you should be able to experiment and see if subtracting the average from each frame prior to integration will help.
Please download RoboRealm again to get that fix.
Thanks again for the great post and helpful images!
STeven.
|
|
|
|
Re:Normalization Fixed
hibou from France [30 posts] |
17 year
|
That's it STeven ! With approx. the same test scene as before I now get the following for 40-30 normalisation of 100 frames
If I first subtract the average, I must change the normalisation to 6-2 for my 100 frames to give
To set the normalisation I just ramp the Ignore Above/Below limits up/down to give the apparently best picture.
This test scene illumination was an extreme case to see how far I could push RoboRealm - surprisingly far, since I can hardly make out anything in the single frame while the integration produces a clearly readable image. It also works almost as well integrating over only 20 frames. I now have this version of RoboRealm running on my beam monitor where the contrast is better, and am very pleased with the improvement.
A big thank you to STeven and RoboRealm ! Alan.
|
|
This Thread
