The Sigma dp2 Quattro is coming soon with the promise of detail approaching that of a 36-megapixel Bayer sensor. Its top layer is full monochrome-grade spatial resolution, with full color information derived by the panchromatic sensitivity of all layers.
I would put it as “full color derivation by virtue of the same physical location for three color layers”, very different from color information separated spatially as in a conventional Bayer matrix sensor (Nikon, Canon, Sony, Fuji), the Bayer approach leading to loss of both spatial and color resolution.
I printed my favorite tulips shot 6 feet wide (~2m) and I deem that it exceeds the quality of my 4X5 film prints of a similar size (they’re on my walls, and the DP Merrill print is unequivocally better in color and contrast and detail, believe it or not).
The foregoing is to make the point that the existing Sigma DP Merrill cameras are already impressive. So to have 19.6 X 3 megapixels* in the new dp2 Quattro in a next-generation sensor vs 14.75 X 3 megapixels is an exciting. I expect the new dp2 Quattro to make cameras like the Canon 5D Mark III look positively challenged for resolution. My prediction is that the 16MP APS-C and Micro Four Thirds cameras will come to be seen as the aging low resolution devices they are.
* Finished RGB or grayscale TIF or JPEG images will be 19.6 megapixels of spatial resolution (vs the DP Merrill 14.75 megapixels). Per-pixel quality (spatial resolution + color discrimination) makes those figures significantly to substantially superior (on a per pixel basis) to Bayer matrix sensor output of similar resolution. In your author’s view, the average advantage is about 50%, but can be as high as double.
The red and green layers
A valid concern might be expressed about about the color resolution versus the prior design. First, all layers are panchromatic; the top layer is blue-heavy, then green-heavy and red-heavy layers. The blue layer in particular captures full spatial and luminance information, peaking in sensitivity to blue/violet.
The red-heavy and green sensitive layers in the Quattro sensor use 2 X 2 sensels, which is half the spatial resolution of the blue-heavy top layer. Note that these 2 X 2 sensels will thus have 4x the photon gathering capability, e.g. about a one stop improvement in noise on average.
The layout leads to the speculation that a derived RGB pixel won’t be full true color as in the previous DP Merrill sensor. But the top layer has full spatial resolution (as if it were monochrome sensor) as well as a broad panchromatic sensitivity. And there is considerable overlap (redundancy) in the signals. Sigma states that signal processing can separate out the true signals for color and luminance.
The images in production cameras will tell the full story, but I deem it likely that the new Quattro sensor design will outperform the previous design on a per-pixel basis.
Conventional Bayer matrix sensors sample two colors per row (G+B or G+R) in a matrix which must be interpreted (guessed) for the actual color. The guessing is known as demosaicing, and it compromises both spatial and color resolution, the impact varying by the subject matter (actual detail and colors striking the sensor). There exist other variants of the Bayer matrix approach but they all similarly compromised, the fundamental compromise being that any given spatial location samples only one color.
The Bayer approach works really well in most cases, but breaks down badly when fine color details exist (e.g. fine blue details on a red or green background color, or similar). Also, Bayer demosaicing can generate color speckling on fine details (spurious color), due to the guesswork of demosaicing; the total result is a significant loss of resolution relative to claimed megapixels.