Alternative color filter array layouts for digital photography

The performance of digital cameras depends not only on the accuracy of methods of restoration of missed color samples (demosaicking) for a given color filter array, but also from spatial configuration of color sensors in the color filter array (CFA) itself. This paper considers three different color filter array (CFA) patterns; the established (2 by 2) Bayer pattern, the 3 by 2 (6-sample) and the 3 by 3 diagonal Bayer CFA. One difficulty in comparing the different schemes is the influence of the demosaicking algorithm on the result. In order to remove this dependence we propose three methods of comparison. They are: (a) measuring widowed averages of colors on large areas (b) visual comparison of interference between regular patterns of images and CFA, and (c) utilization of one layer neural networks to build demosaicking algorithm for selected color filter arrays. A substantial image database comprising 1338 images has been used to experimentally validate the different patterns.