Why are there two green sites in a Bayer block?

That is the pattern's sampling arrangement; the reconstruction must account for its unequal channel sampling.

Does every camera need Bayer demosaicking?

No. Sensor and filter designs differ, and some do not use a Bayer color filter array.

Can interpolation create artifacts?

Yes. Incorrect estimates around edges or fine textures can produce false colors and other visible errors.

AI Glossary / Computer vision and multimodal AI / Advanced

Demosaicking

Definition

Demosaicking reconstructs full-color image values from measurements captured through a color filter array. In a typical Bayer sensor arrangement, each photosite measures through one red, green or blue filter, so the other color components at that location must be estimated. The process uses nearby measurements and, in more advanced methods, relationships among color channels and image structures.

What it means

A raw color-filter-array image does not contain an independently measured red, green and blue value at every location. Demosaicking fills those missing components computationally. In a repeating Bayer block, two sites have green filters, one red and one blue. This arrangement is common but not universal; other filter patterns and sensor designs need different processing. Reconstructed detail depends on the measurements, scene and method. The output's full RGB grid should not be mistaken for three fully observed channels at every original sensor position.

How it works

A basic method interpolates missing channels from nearby samples measured through the relevant filters. More sophisticated approaches adapt to edges, exploit cross-channel relationships or estimate residual detail to reduce artifacts. Fine repetitive textures are challenging because the scene may contain detail the sampling pattern cannot represent reliably. Noise, optical blur and subsequent sharpening also influence appearance. Demosaicking is one stage of an imaging pipeline and should be distinguished from white balance, color correction and tone mapping, even when a camera or raw processor performs these operations together.

Three practical examples

A raw-photo editor reconstructs color from a Bayer capture and inspects fabric patterns for false colors introduced near fine repeating detail.
A camera engineer compares demosaicking methods using controlled targets, looking at edge artifacts as well as processing cost.
A machine-vision team checks whether its analysis expects raw mosaic data or reconstructed color, so the input representation matches the processing assumptions.

Think of it this way

Imagine a checkerboard where each square reports only one ingredient of a color recipe. To prepare a complete recipe at every square, you borrow clues from neighboring squares and from how colors change across boundaries. If a sharp boundary passes between them, blindly averaging can mix information from opposite sides. The analogy explains interpolation and edge awareness. It does not imply that the missing components are known exactly or that every sensor uses a simple checkerboard arrangement.

Worked example

In an original simplified bilinear example, a red-filter site records a red value of 100. Its four immediate green-filter neighbors measure 80, 84, 88 and 92. Their average is 86. Four diagonal blue-filter neighbors measure 60, 64, 68 and 72, whose average is 66. The reconstructed RGB value at the red site is therefore (100,86,66) under these stipulated interpolation rules. Only the red component was measured at that location. The other two are estimates, and a nearby edge could make the averages misleading. This calculation assumes compatible linear values and omits boundary handling, noise correction and later color processing.

Illustrative case study

A fictional online retailer notices colored fringes in photographs of tightly woven jackets. The imaging team compares raw-processing methods using the same captures and checks detail at both full resolution and the website's display size. One method reduces false color but softens texture, while another preserves apparent detail with more artifacts. The team selects settings for its presentation needs and retains the raw files for future processing. This scenario illustrates a reconstruction tradeoff, not a guarantee that a particular method can recover all fabric detail or that visually sharper output contains more correct information.

Common misconceptions and mistakes

Demosaicking is not simply enlarging an image, although both can involve interpolation. It also does not recover deliberately obscured content from a privacy mosaic. Another mistake is applying the wrong filter pattern or ignoring the sensor's orientation and crop origin. Those errors can produce incorrect colors even if the interpolation code runs successfully. Fine detail can create zipper-like edges or false color, and sharpening may emphasize these artifacts. A full-color output file does not reveal which channel values were measured directly and which were reconstructed.

Business use

The concept matters when choosing or validating cameras, raw-processing workflows and vision-system inputs. Compare methods on representative subjects rather than only smooth test images. A business photographing textiles may care about false color in repeating patterns, while a measurement workflow may prioritize stable edges and consistent processing. Record the raw format and reconstruction settings when reproducibility matters. Validate downstream models with the same kind of reconstructed input they will encounter in use; changing the imaging pipeline can change image statistics even when the scene and camera remain the same.

Check your understanding

Which component was directly measured at the example's red site?

Answer: The red value of 100.

Why were green and blue estimated?

Answer: Their filters were at neighboring sites, not that location.

Does obtaining three output values prove all three were observed there?

Answer: No; reconstruction supplied the missing components.

Related terms

Color filter array (CFA)
Bayer pattern
Aliasing (factorial experiments)
Color filter array
interpolation
aliasing
raw image
color correction

Learn next

Sketch a Bayer block and identify the missing components at each site. Compare bilinear interpolation with an edge-aware method on a diagonal boundary. Then inspect a fine repeating pattern to understand why sampling limitations can remain even with sophisticated reconstruction.

Bilinear interpolation

Sources

Getreuer, Malvar-He-Cutler Linear Image Demosaicking: https://www.ipol.im/pub/art/2011/g_mhcd/revisions/2011-08-14/g_mhcd.htm Duran and Buades, A Demosaicking Algorithm with Adaptive Inter-Channel Correlation: https://www.ipol.im/pub/art/2015/145/ The channel values, recipe analogy and retail-photography case are original educational illustrations. The worked example describes basic interpolation rather than either cited algorithm in full.

Previous: Demonic nondeterminism Next: Dempster-Shafer theory