What are color artifacting/aliasing errors? Why do they occur? What can we do about them?
The fact behind why single pass cameras exhibit this tendency is because we can't place a red and a green and a blue filter at every pixel site (that would be N.D., not color). We can place a red, or a green, or a blue pixel at every site, which gives us the ability to shoot a color picture in one exposure, if we "develop" that picture. What does development do? Let's explore some fun facts (!!!) about how digital cameras make color, and then we'll learn why these red and blue color artifacts happen (this is color artifacting, sometimes called a christmas tree effect).
Development manufactures the two colors not physically present at every pixel site. Since we can only put one colorant, we must invent the other colors, we use the term "interpolate" to describe this mathematical routine. Each digital camera company does their own unique method of making this invention. For example, each green pixel needs to have a red and a blue interpolated at the green pixel site. After development you can look at the red, green, and blue channels and see the results of the interpolation, even though we didn't have those pixels to capture the shot.
The silver and gold pixels are the result of interpolation errors. When we develop, we look at a large number of pixels to interpolate (we have 4 million to eval on the S2, 6 million to eval on the S3). When we invent the two missing channels we like to evaluate a large area surrounding each pixel site, to see what color they are, because that helps us invent those channels correctly.
The first thing we have to do is decide what colors are focussed by the lens on each and every pixel. Let's consider the green pixels for a moment. If green pixels have a large electron count (that's how we count light energy, we're actually in the business of counting electrons) we can make two assumptions, either green light landed on them or, alot of light energy of indeterminant color landed on them. If, the red pixels adjacent to the green ones have a low electron count, we can make a reasonable guess that green light landed on the green pixels because we know green light passes thru green filtration without too much absorption, while red filtration does a pretty good job of absorbing green wavelengths of light. Next, we'll look at the adjacent blue pixels; if they had an electon count between the green and red counts, we're in a postion to strongly assume that green light landed on the green pixels because our blue filtration passes a known amout of green light energy. This assumption would allow us to make the appropriate red, green, and blue channels for a green reflectance.
This process is how single pass cameras make color when they use a ccd whose colorant is applied individually to each pixel site. We use the industry standard Bayer pattern, a pattern that contains 2 green pixels, one red, and one blue, in a bicubic interpolation area. Human visual perception prefers more resolution (luminance) than color (chrominance) because we have more rods than cones in our retinas. Because of this preference, the Bayer pattern has more green pixels than red or blue (this fact is quite pertinant to the blue and red color artifacts), that's because green pixels make the greatest contribution to the luminance channel. Our eyes (and brains) will adjust for color but not for resolution; it's why we don't strongly object to flourescent light when we see it, but will object to a photograph when it's soft, but color correct. That same photographic image can be off in its color but if it sharply focussed, we might not complain too much.
So why do these color artifacts occur? Partly because the red and blue are what we call undersampled as compared to the green pixels. And, partly because of what we point our camera at. If the lens resolves sharply enough, adjacent pixels could be an entirely different color if the reflectance is one of specular highlights or maybe a fine fabric texture. If, across the area of several pixels we have several different colors, we get the color artifacts (which we will need to fix in a post process). The specular nature of the silver or gold jewlry are due either to the lighting or to the surface of the jewlry (facets on a gemstone for example).
Post processing for this type of artifacting includes MV's Fast filter, the Adaptive filter and on a selection basis, the RGB Fixer. There is an order to consider when applying filtration for stubborn reflectances. I usually include None for filtration and decide after capture what to do. If nothing is the appropriate choice, I do nothing. I typically choose Adaptive (very little loss of resolution for this filtration) and, if that does not deliver acceptable results I filter secondarily with the Fast filter. I've been generally happy with this scenario but when it does not give acceptable results I develop the MEGA file a second time and run the RGB Fixer. I develop a second time because the RGB Fixer needs data that is sometimes removed by the Fast and Adaptive filters.