PDA

View Full Version : Questions about the noise reduction feature of ICCapture



acanicio
March 11, 2008, 23:30:38
Hello,

I'm curious about how the noise reduction feature of ICCapture works.
If I understand correctly the explanations in the manual, it seems that it averages from 1 to 32 frames in real time.

However I have two questions :

1 - When using 32 frames, with a 30FPS framerate, I should get about 1 frame every second. How come the resulting frame rate is about 10 FPS ?
Is it a "slipping" average of frames ?

2 - When averaging frames, I should notice at least some smearing, since all the frames must have a different timestamp. But I have not been able to notice ANY smearing, however fast I move the camera. This is really curious...

Thank you for sharing the secret... :-)

Best regards
Axel

Stefan Geissler
March 12, 2008, 12:48:06
Axel,


You found an error, that has been in IC Capture for three years. You are the first one you encountered it. It will be fixed in one of the next versions. (But this should be a secret :-) )

acanicio
March 12, 2008, 19:27:15
Dear Stefan,


Axel,

You found an error, that has been in IC Capture for three years. You are the first one you encountered it. It will be fixed in one of the next versions. (But this should be a secret :-) )

Oooooops.... sorry about that :-))))
It's the fault of my "twisted tweaker" character...

Best regards
Axel

Stefan Geissler
March 13, 2008, 10:20:37
Axel,

Do not worry about the "twisted tweaker" character. We appreciate bug reports and hope to create a better product. The programmer already works on these problems.

martinel
April 22, 2008, 18:06:21
Hello,

I also noticed that most of the values for noise-reduction in IC Capture do not work as expected.
The setting "4 images" does create the expected number of images with lower noise, though.

It would be nice to have the full flexibility of averaging 2,3,4,5 ... images, instead of just the current power of 2 values, but making the existing settings work is more important, of course.

Regards, Martin

Stefan Geissler
April 23, 2008, 09:20:40
Hi Martin,

This is a matter of efficiency. If you add 5 frames, then you have to divide each byte of the image by 5. If you only use powers of 2, then the division can be done with only one CPU cycle with a simple shift right command. The division needs more than one CPU cycle. When the Denoise filter was created, this was real advantage of time.

martinel
April 23, 2008, 11:30:21
Hi Stefan,

Oooops, thank you for that information! That is a problem. :eek:

This numerical approach is quite problematic for astronomical applications, as it destroys valuable information in the LSBs. For astronomical data, where the light-sources often are very weak we should avoid to destroy any information and the LSBs often hold such information.

I would have expected the noise-reduction to really average the brightness values, by adding them all first (in an int or short) and then dividing. This approach will be slower, but the speed-optimized approach might loose a lot of the information we are actually looking for.

I will have to evaluate this problem, the current implementation might make the feature useless for my application.

Best regards,
Martin




Hi Martin,

This is a matter of efficiency. If you add 5 frames, then you have to divide each byte of the image by 5. If you only use powers of 2, then the division can be done with only one CPU cycle with a simple shift right command. The division needs more than one CPU cycle. When the Denoise filter was created, this was real advantage of time.

Stefan Geissler
April 23, 2008, 12:48:48
Martin,

The adding is done first and then the shifting (dividing) is done. Information is not lost.