Please click "original" below to see the crops at 100% (or 1:1). Investigating resolution and leaked response of the astronomically important deep red H-alpha in G and B pixels, in a stock Canon 1Ds and in a Hutech modified 20D. This 20D has only clear glass with anti-reflection coatings over its sensor, no anti-aliaising filter. The H-alpha filter is by Astronomik, passband nominally 13nm. The side columns show response in visible light (1Ds unfiltered, 20D used with a UV/IR Blocker). All frames had their contrast maximally stretched between their white and black points. The full frames have been shrunk to their respective sizes, i.e. demonstrating the infamous 1.6x crop factor of the 20D, with the yellow highlighting the 200x100 pixel crops. Since the pixel sizes differ between the two cameras, the images look as if they are at different enlargements, even though they were taken with the same lens at the same distance. Canon 100mm/2.8macro USM at f8, autofocused. Leaked response of H-alpha into G and B is practically nonexistent in a 1Ds (< 1% of the R response) but it would appear that Canon is using different dyes in the Bayer array of the 20D. R:G:B response for Ha is roughly 100:5:13 respectively. Conclusion? When shooting H-alpha in a modded 20D you may wish to use the Luminance Channel, rather than the Red Channel for further processing. It seems to carry some boost in resolution from the leakage, compared to the Red Channel. Even though the Green Channel is sharper, it's also much more noisy. In practice however, blurring from imperfect tracking and seeing conditions is often greater than the loss of resolution in H-alpha, and one hardly ever displays astro-images from these multi-megapixel chips at 1:1 anyway, even after cropping. All the above have had no sharpening applied, neither in-camera nor in post-processing.
For other stuff peruse my primary website: http://www.samirkharusi.net/