A Structural Approach to Improved Colour Filter Array Demosaicking for the Bayer Pattern

J.S.J. Li and S. Randhawa (Australia)


Bayer pattern, CFA demosaicking, structural approach


According to the Bayer pattern, a colour filter array (CFA) image contains only fifty percent green pixels, and twenty-five percent red and blue pixels. It has been shown that the missing colour values can be extrapolated to produce an accurate estimate. For high accuracy in extrapolation for a missing colour, it is desirable to include its nearest neighbouring values in the CFA image. Based on the structure of the Bayer pattern, the nearest neighbours of a missing green pixel are located to its immediate top, bottom, left and right. However for the red (or blue) plane, half of the number of pixels have its nearest neighbouring values situated on the diagonals. Accordingly, two different edge orientation maps, one for the horizontal and vertical edges and one for the diagonal edges, have been proposed to identify the direction for extrapolation. This has an advantage over other CFA methods which only use one edge indication map for all colours. It has been shown that our method outperforms the existing demosaicking methods in both the visual and quantitative analysis.

Important Links:

Go Back