The first option would be to design a machine like the Palette. However, rather than splicing multiple distinct filament colors, we add more melt chambers (the current Palette has only one). The device would support five input colors (white, cyan, magenta, yellow, and black). In that way, it would work the same as taking a CMYK or RGB value to Home Depot and saying, "Make me this color of paint." They start with either black or white paint, and then they add small amounts of the other colors to get the desired result. The machine keeps all five input colors in a melted state, in something like a hot end. The steppers push cold filament through the melt zones, which all converge on a central 1.75mm bore, which has some kind of mixing apparatus, which could be a rotating paddle, or passive fins on the inside that introduce enough turbulence to efficiently mix the colors. This could be implemented on an external device, such as the Palette, or it could be integrated into the printer itself, although that would make it rather top-heavy.
The second option would be a modification of the pellet feed system in SeeMeCNC's Part Daddy delta, which is freaking huge, and which has a hopper that's filled with plastic pellets rather than using spools. It melts the plastic right at the hot end. As above, this would be outfitted with five hoppers (WCMYK) and it would mix the pellets in the correct proportions. Obviously, this would be for super huge printers like the Part Daddy.
The biggest difficulty I see is in allowing rapid color changes. To print something like a photograph would require extremely fine control of the mixing. That would mean a very small mixing chamber. I don't know how short you could realistically have an individual color segment without bleeding. In fact, the current Palette machines assume you'll be drawing a purge wall. If we use converging melt chambers, it might be possible to distribute a number of tiny ports into the main chamber. Rather than having five inputs, there might be 10 to 30 (some multiple of five), all evenly spaced, so no one area of the main melt chamber would be overly biased towards one color. This could prove very difficult to manufacture as a single piece - it would probably require a selective laser sintering machine, printing with titanium powder. Another possibility would be to have each color's melt chamber fork off into several very thin copper tubes wrapped with ni-chrome wire to keep the plastic molten, and these would connect to multiple tiny ports in the main melt chamber.
The third possibility is to adapt inkjet technology. The dye injection method doesn't do this, so AFAIK, those patents wouldn't apply. There's already a project that lets you drive inkjet print heads with an Arduino. The idea would be to put down a layer with FDM, and then return and print the colors on the edges of that layer. This seems to me like it could be a bit messy and error-prone, and I don't know whether it's already covered by a patent. If no one has patented it yet, let this post be the thing that stops them from doing it.
Obviously, an inkjet setup would be bulky, and might be better for printers with a fixed toolhead and moving platform, e.g. polar printers, and deltas with a moving print surface.Apple has already patented a technology that would go over the print after it's already done, and paint it using some kind of contour-following mechanism.
