Creative Color Infrared

Seeing what would happen if an identical Infrared & Visible image could be "fused" in Photoshop, using several methods involving assigning color channels, I came up with what might be the most interesting color scheme. The image below is the result of assigning the Infrared (850nm) image as the Red channel, while red was assigned to Green, and finally, Blue & Green were both assigned as the Blue channel.

side note: it would be much easier for artifacts such as ghosting to be invisible, by using two identical cameras, one to image the IR and the other to image a normal image. I was at one point even able to use luminance from Infrared along with color from the normal image, as well as another way of merging color channels, to be able to make for black skies along with reddish/orange foliage, which I personally found really great, although I've forgotten exactly which technique I used for that. Anyone ever try these Infrared methods before? Will be interesting if anyone also does this.

MacM545 wrote:

Seeing what would happen if an identical Infrared & Visible image could be "fused" in Photoshop, using several methods involving assigning color channels, I came up with what might be the most interesting color scheme. The image below is the result of assigning the Infrared (850nm) image as the Red channel, while red was assigned to Green, and finally, Blue & Green were both assigned as the Blue channel.

All sorts of false color assignments are possible, and there are standards for this, with LandSat data being the most common source. For example, see False-Color Composite with Multiple Images Tutorial . Of course, you can also do LUT-based mappings that are much more complex.

My easiest way to do this is to start with one of those few cameras that has a 4-color CFA rather than the usual RGB. For example, some early Canons had CMYE and the Sony F828 has RGBE (and supports NightShot, so you don't need to modify it to remove the NIR-blocking filter). It's rarely discussed outside of the raw conversion software development community, but some more modern cameras actually have two slightly different green channels -- look at the source code for dcraw .

Perhaps surprisingly, four color channels gives you four linear equations in three unknowns (mapping four CFA colors into R, G, and B) -- an overspecified system -- but it naturally isn't overspecified if you treat it as four equations in four unknowns! Thus, assuming the sensor sees NIR, you can directly solve for R, G, B, NIR. The biggest problem is that the approximate signal levels for the four channels might not match, in which case the weaker channels will suffer greater gain factors, increasing noise. For example, this happened when I solved for R, G, B, NIR using a stock PowerShot G1 (work I did way back in 2001 ); the NIR channel was MUCH darker than RGB. However, a visible-light-blocking filter added to an unmodified camera can equalize the sensitivities... although that gives about 7 stops poorer sensitivity than removing the NIR-blocking filter from the sensor stack.