Digital cameras overlay a Color Filter Array (CFA) over the sensor which sub-samples the color information of the scene. The full color image is then recovered using a class of algorithms known as demosaicing. The color filters used for CFA are predominantly Red, Green and Blue. In this article we propose an algorithm based on Linear Minimum Mean Square Error (LMMSE) which can demosaic images from any color (linear combination of Red, Green and Blue) filter arranged in any order in the CFA. We also propose optimum CFAs based on combination of RGB colors which outperform the state of art CFAs in image reproduction with less computational complexity.