Moire removal for Debayer algorithms
It's not enough to get high PSNR from debayer algorithm to be sure that image quality is good. Quite often one can run into situation when algorithm with really good PSNR gives false colors and we see moire artifacts over the image. To clear up the case, we consider the following debayer algorithms: bilinear (AN3 from Intel IPP library and NVIDIA NPP), AHD (Adaptive Homogenuity Direction, from Intel IPP), HQLI (High Quality Linear Interpolation, which you can find at OpenCV and Fastvideo SDK), DFPD (Directional Filtering and a Posteriori Decision, from Fastvideo SDK), DFPD_R (DFPD with color correction stage, from Fastvideo SDK), Amaze (from Raw Therapee) and MG (Multiple Gradients, from Fastvideo SDK).
At the following table we show the source image and the results of debayering according to corresponding color interpolation algorithm together with PSNR value. To see each image with more details, please click on the image to download it at full resolution.
It's quite clear to see false colours on interpolated images and to compare artifacts that we get after debayer. The best PSNR shows Amaze algorithm due to better interpolation of high frequency details on the image. Though, if we consider just inner part of the source image (these are mostly low frequency details), then the best PSNR will be at MG debayer algorithm.
Quality measurements for different debayer algorithms (with or without JPEG compression) you can find here.