Algorithms Demosaicing

1 algorithms

1.1 simple interpolation
1.2 pixel correlation within image
1.3 video super-resolution/demosaicing
1.4 tradeoffs

algorithms
simple interpolation

these algorithms examples of multivariate interpolation on uniform grid, using relatively straightforward mathematical operations on nearby instances of same color component. simplest method nearest-neighbor interpolation copies adjacent pixel of same color channel. unsuitable application quality matters, can useful generating previews given limited computational resources. simple method bilinear interpolation, whereby red value of non-red pixel computed average of 2 or 4 adjacent red pixels, , blue , green. more complex methods interpolate independently within each color plane include bicubic interpolation, spline interpolation, , lanczos resampling.

although these methods can obtain results in homogenous image regions, prone severe demosaicing artifacts in regions edges , details when used pure-color cfas. however, linear interpolation can obtain results when combined spatio-spectral (panchromatic) cfa. 1 exploit simple formation models of images demosaicing. in natural images within same segment, ratio of colors should preserved. fact exploited in image sensitive interpolation demosaicing.

pixel correlation within image

more sophisticated demosaicing algorithms exploit spatial and/or spectral correlation of pixels within color image. spatial correlation tendency of pixels assume similar color values within small homogeneous region of image. spectral correlation dependency between pixel values of different color planes in small image region.

these algorithms include:

variable number of gradients interpolation computes gradients near pixel of interest , uses lower gradients (representing smoother , more similar parts of image) make estimate. used in first versions of dcraw, , suffers color artifacts.
pixel grouping uses assumptions natural scenery in making estimates. has fewer color artifacts on natural images variable number of gradients method; introduced in dcraw rel. 8.71 patterned pixel grouping .
adaptive homogeneity-directed interpolation selects direction of interpolation maximize homogeneity metric, typically minimizing color artifacts. has been implemented in recent versions of dcraw.

video super-resolution/demosaicing

it has been shown super-resolution , demosaicing 2 faces of same problem , reasonable address them in unified context. note both these problems address aliasing issue. therefore, in case of video (multi-frame) reconstruction, joint super-resolution , demosaicing approach provides optimal solution.

tradeoffs

some methods may produce better results natural scenes, , printed material, instance. reflects inherent problem in estimating pixels not know certain. naturally, there ubiquitous tradeoff of speed versus quality of estimation.