A Kodak image, captured by Sony RGBW CFA (left) and by IA RGBW CFA (right). The Sony image suffers from more false coloring.
More images available here. Test conditions for generating the image set is outlined in the included Readme.txt file.
The popular Bayer Color Filter Array (CFA) captures high quality images in bright light, but poor images in low light. Its color filters transmit only 25% of the incident light, which severely degrades its sensitivity.
This motivated Kodak to develop RGBW CFAs that transmit over 50% of the incident light, thereby greatly improving its luminance SNR. The trade off is decreased chrominance SNR which is not a big concern since chrominance can be effectively denoised. Unfortunately, all RGBW CFAs developed so far have suffered from mosaicking-demosaicking issues such false coloring and aliasing.
The mosaicking−demosaicking process is a tougher problem for RGBW CFAs than it is for RGB CFAs, such as Bayer, because the extra W color plane has to be shoehorned into the mosaic without significantly reducing the resolution of the R, G, B and W color planes. What’s worse, W has to be sampled at high density in order to maximize sensitivity. This leaves very few R, G, B pixels and thus complicates the demosaicking problem.
Several years after the initial Kodak RGBW design, Aptina developed the Clarity+ CFA. This CFA overcame false color, moire and luminance resolution problems at the expense of color fidelity by replacing the Green filter in the Bayer pattern with a White/Clear filter. This RWB CFA captures R, B directly, and with high fidelity, but sacrifices G fidelity by inferring it as G = W−R−B.
An alternative dual camera architecture has been proposed that uses a monochrome camera to capture a clean luminance image and a color camera to capture noisier color image. The chrominance signal extracted from the color camera is denoised and combined with the clean luminance from the monochrome camera, much as it is done by RGBW systems.
While the dual camera system overcomes the mosaicking-demosaicking problem of RGBW CFAs it introduces the problem of fusing the two images, which is a compute intensive operation that fails for subjects occluded in one image but not the other. Three camera systems have been proposed for the occlusion problem, albeit at an even greater cost. Other features provided by dual camera systems, such as greater sensor area and depth sensing can be performed equally well with dual RGBW cameras.
The salient points of IA's RGBW CFA are: