Impact of Photometric Space Linearity on Demosaicing Image Quality

Capturing visual scenes in color by using image sensors at a lower cost has resulted in the widespread use of color filter arrays (CFAs) where spectral sampling is used to trade off image fidelity at the expense of image sensors and hardware complexities. Obtaining color images from quincunx sampled pixel values, where each measured pixel value has only one color available, by estimating the missing remaining color components at each pixel is called demosaicing (aka demosaicking, demosaicing) or CFA interpolation. In this paper, we discuss the use of linear and nonlinear photometric spaces in demosaicing algorithms and objectively analyze their impact on the image quality with CPSNR, S-CIELAB, and SSIM metrics using Kodak PhotoCD and McMaster image data sets.