Bayer De-mosaicing

In literature different algorithms for interpolation of an color image from a Bayer Pattern image (Bayer demosaicing) exist. In this section we give several examples of such algorithms. In the following table they are summarized:

Algorithm Description Example
“Nearest Neighbor filter" according to [Ada95] 'Nearest Neighbor De-Mosaicing'
“Bayer3x3filter" bilinear interpolation; based on VisualApplets operator BAYER3x3Linear 'Bayer 3x3 De-Mosaicing'
“Bayer3x3filter" with white balancing bilinear interpolation; based on VisualApplets operator BAYER3x3Linear with additional white balancing 'Bayer 3x3 De-Mosacing with White Balancing'
“Bayer5x5filter" bilinear interpolation; based on VisualApplets operator BAYER5x5Linear 'Bayer 5x5 De-Mosaicing'
“Bayer5x5filter" with white balancing bilinear interpolation; based on VisualApplets operator BAYER5x5Linear with additional white balancing 'Bayer 5x5 De-Mosacing with White Balancing'
"Laplace filter" edge sensitive 'Edge Sensitive Bayer De-Mosaicing Algorithm'
"original Laroche filter" according to [Lar94] 'Bayer De-Mosaicing Algorithm According to Laroche'
"modified Laroche filter" according to [Lar94]; modified 'Modified Laroche Bayer De-Mosaicing Algorithm '
Bilinear Bayer Red/BlueFollowedByGreen GreenFollowedByBlue/Red according to [Bas13] 'Bayer DeMosaicing For Bilinear LineScan Cameras with Color Pattern Red/BlueFollowedByGreen GreenFollowedByBlue/Red '
Bilinear Bayer Red/BlueFollowedByBlue/Red GreenFollowedByGreen 'Bayer DeMosaicing For Bilinear LineScan Cameras with Color Pattern RedFollowedByBlue GreenFollowedByGreen '

Table 10. List of Bayer Demosaicing Examples


The algorithms named above are explained in detail in the corresponding sections. Here we give a short overview on the qualitative results obtained with the different demosaicing methods for an artificial test image (Figure 230, 'Artificial test image').

Artificial test image

Figure 230. Artificial test image


The following figures show closer details on single structures. From left to right for: a) the "original Laroche filter" [Lar94], b) the "modified Laroche filter", c) the "Laplace filter", d) the “Bayer5x5filter", e) the nearest neighbor interpolation method and for f) the “Bayer3x3filter":

Straight edge

Figure 231. Straight edge


Diagonal edge

Figure 232. Diagonal edge


Curved edge

Figure 233. Curved edge


Periodic structure

Figure 234. Periodic structure


For the artificial test image the "original Laroche filter" shows the best subjective result according to zipper- and false color effects. The "modified Laroche filter" has light zipper effects and is comparable to the "Laplace filter". For the “Laplace filter" in addition many false color pixel appear at curved and diagonal edges. The “Bayer5x5filter" shows strong zipper effects, whereas the “Nearest Neighbor filter" shows strong false color effects. For straight edges strong zipper effects appear here. All edges appear blurry with the "Bayer3x3filter".

Attention: The "original Laroche filter" is not always the best solution for Bayer demosaicing! It has no or very few zipper effects for an artificial test image but may have those for a noisy Bayer RAW image. During research we found, that the filtering algorithm is influenced by the noise.The "modified Laroche" filter" has zipper effects for the artificial test image but only light zipper effects for a noisy Bayer RAW image. Here the filtering algorithm is less than the "original Laroche filter" filter influenced by the noise. Also the "Laplace filter" has zipper effects for the artificial test image but no zipper effects for a noisy Bayer test image. False color effects appear here in addition. In a consequence you have to choose, which Bayer demosaicing algorithm is the best for your special purpose.