We have all read about the Bayer array and its supposed destruction of resolution. But does it really happen? Make your own conclusions. I used the sharpest lens I own, the Canon 100mm/2.8macro USM, at its sharpest f-stop (f8), and shot through various filters trying to demonstrate the resolution loss. Camera is the Hutech-modded 20D with clear glass over the sensor, i.e. no low-pass anti-aliaising filter as in the stock version. First attempts were thwarted because what looks like high contrast lines in visible light suddenly lose most of the contrast in Infra Red. Consequently I had to reshoot using a fly screen. For all the photos I autofocused through the respective filter(s) at that piece of duct tape stuck in the center of the viewfinder frame. On the extreme left you see an uncropped full frame. Now click on "original" below. The crops are now 1:1, at 100% display.
No Filter = visible light + Near IR
Ha = Astronomik H-alpha CCD filter
UHC = Astronomik UHC filter, passes through Hb, OIII, Ha, SII + "some" IR
R72 = Hoya R72, blocks visible light, passes through 720nm and longer (glass, at front of lens)
89B = Wratten gel, blocks visible light, passes 720nm and longer, as a pop-in between lens and mirror box of the camera
UI = Hutech IDAS Type 1 UV/IR Blocker
My conclusions? Much ado about very little. The Ha filter is expected to show the most detereoration in resolution, owing to only one quarter of the pixels being Red. Does it show this to a disturbing extent? Your call. Since visual acuity (mine anyway) seems diminished for monochromatic Blue or Red images, I decided to display unsharpened B&W versions of the whole lot. All had the brightness histogram stretched fully between the black point and white point. The lens is sharpest, most contrasty, when it is used as it was designed for, for visible light only, with the UV/IR Blocker, top right. So why no significant loss in resolution in Ha, something I had noted in an unmodded 1Ds astrophoto? On an unmodified 1Ds the spill-over/leakage response in G and B pixels is <1% of the R values. Is there leaked respnse of Ha in a modded 20D? See next slide.
