View Single Post
Old 30-04-2012, 01:07 PM
PRejto's Avatar
PRejto (Peter)
Registered User

PRejto is offline
Join Date: Jan 2011
Location: Sydney, Australia
Posts: 1,296
Over the past week I have had quite a discussion over on the Imaging Source Forum, and private correspondance with Emil (AS!2). Several interesting points have emerged that might be of interest. It was new material to me...

1. Imaging Source now admits that the DeBayering pattern within IC Capture is incorrectly labled. Within the program one must turn on "BG" to see the correct colors pre-capture. The correct setting is "GB" for the DBK21AU618 and is determined by the chip.
2. Imaging Source will not correct the error because they claim that all other programs designed to work around IC Capture would then fail.

More here:

This explains the inconsistency I've long experienced in selecting the correct DeBayer pattern outside of IC Capture (such as AS!2, Firecapture, etc)

3. The "GB" setting will only work correctly if the capture is full frame. According to Imaging Source, using ROI will in 3 cases out of 4 cause a shift in the DeBayer pattern needed to decode. Thus, if you capture using ROI you will need to guess what the correct pattern is by trial/error. My "blue" mars was caused by selecting a pattern I "knew" to be correct but wasn't because I had used ROI to capture.

4. There is no coding within Y800/Y800 video to tell any program what the correct pattern is. According to Emil the software tries to guess based on the fact that usually the blue channel has the most noise. With small image captures AS!2 can get this incorrect. Anyway, Emil said I could quote him and he explains it better:

"There is no way to know for sure which grid is used in a recording, this information is never saved. AS!2 tries to guess the color information from the statistics of the pixels (the blue pixels are often noisier than the green and red pixels, and the two recurring green pixels should have a very similar noise level). When there are less reliable statistics available (less pixels), it becomes easier to guess a wrong bayer pattern (it might be by chance that the blue pixels are in fact a bit less noisy, than the red pixels, resulting in a switched red/blue channel). Cropping can influence the pattern, if you crop the first two lines and columns of the previous pattern, you'll end up with:

5. My Mars images were helped a lot by not using the generic IR/UV filter I got from Bintel. The filter by TruTek works much better and I get better range out of the red balance control.
Reply With Quote