QUOTE (tedstryk @ Dec 13 2004, 03:45 PM)
Perhaps mask could be created using pancam data for the areas covered and interpolating the rest.
I've been wrestling with this same problem for about 3 months now, with only limited success. My first attempts were to build a navcam-equivalent out of pancam (using multiple pancam filters in a certain ratio to create a single black and white image that is a replica of the navcam) and corresponding the brightness values of that pseudonavcam to the color pancam to build a gradient. Apply that gradient to the brightness values of the navcam and voila! it looks really bad...
The problem is that when I take a pancam/navcam pair with two seperate sets of features in a specific image that have distinct hues and try it, you end up with a gradient that represents the average hue of both (which in practice just means that both hues are then wrong). And almost all the navcams of interest have at least 2 if not 3 distinct hues (ground, rock, sky).
My current attempt is to pull out by image segmentation those 3 distinct areas, and attempt the same method in each of the 3 areas individually.
Here's a little tool I built to figure out if this method is going to work...
This is just the graph of the navcam to pancam brightness correspondance in the green channel of a spirit image with only ground and rocks in it (no sky). The X axis is the brightness of the navcam, the Y axis is the brightness of the green channel in the color pancam and the brightness of the pixel in the graph represents the number of pixels that correspond.
There are two distinct bands, one corresponding to the brightness in green of the rocks given a certain navcam brightness, and the other band is the ground/sand/dust. Since these bands are fairly tight, if I can find a good way to pull out 'just the rocks', 'just the sky' and 'everything else' from the pancams (which isn't that big a problem, since I can work from the color images which helps pull out those distinctive areas) and navcams (which is a huge problem since the line between sand/dust covered rock and the exposed rockface itself is less than obivous in most navcams) in an automated fashion, i'll be on the road to a pretty decent navcam colorization.