In most digital cameras, color filter array (CFA) is used for sampling only one color value for each pixel on CCD. So in raw images captured by digital cameras, every pixel has only on color in R (red), G (green) and B (blue). The other two color values need to be interpolated. The interpolation process is commonly known as demosaicking....[ Read more ]

In most digital cameras, color filter array (CFA) is used for sampling only one color value for each pixel on CCD. So in raw images captured by digital cameras, every pixel has only on color in R (red), G (green) and B (blue). The other two color values need to be interpolated. The interpolation process is commonly known as demosaicking.

In this paper, the relationships inter and intra color channels are analyzed, and we will discuss relationships among the sub images in frequency domain. Basing on the features, two frequency merging (FM) demosaicking algorithm are proposed, which merge raw sub images in frequency domain with different methods.

The first one is using statical result of distribution of distortions to estimate the raw sub image with minimum distortion to target sub image for every pixel in frequency domain. Values of optimal raw sub images are picked and assigned to target sub images directly.

The second one is merging the four raw sub image using linear combination. The weights are calculated with LMMSE method. So one estimatation image is needed for calculating minimum mean square errors.

We give experiment results for both algorithms. Both algorithms have good performance at PSNR, especially LMMSE FM method. In visual quality, FM methods could also eliminate color artifact obviously.