Why are there two green sites in each common unit?

The design allocates denser sampling to the channel that strongly supports brightness detail, while red and blue samples supply other color information.

Does every digital camera use a Bayer pattern?

No. Other color-filter layouts and sensor arrangements exist.

Can I choose any Bayer order when opening raw data?

No. The interpretation must match the actual sensor arrangement and image orientation. A wrong order can produce severe color and edge errors.

AI Glossary / Computer vision and multimodal AI / Intermediate

Bayer pattern

Definition

A Bayer pattern is a repeating arrangement of red, green and blue color filters over a single image sensor. Its familiar two-by-two unit contains two green-filtered sites, one red-filtered site and one blue-filtered site, enabling color capture through a spatial mosaic of measurements.

What it means

A photosite under one filter records light through that filter rather than separately measuring complete red, green and blue values. The sensor therefore produces an interleaved pattern of color samples. Demosaicking estimates the missing color components needed to construct a full-color image.

The pattern uses more green samples than red or blue samples, supporting the representation of brightness detail. This does not mean that the camera sees green objects twice as accurately in every respect or that its final image should look green.

RGGB is one common way to name the two-by-two arrangement, but other orderings arise with sensor orientation and the chosen starting position. Correct interpretation of the pattern is essential when processing raw data.

How it works

Light passes through the color-filter mosaic before it reaches the corresponding sensor sites. A readout records each site's measurement along with the spatial arrangement needed to interpret it. The filters have real spectral responses, not idealized single-wavelength selectivity.

The processing pipeline handles operations such as black-level correction, gain adjustment, demosaicking and color conversion according to its design. Demosaicking uses relationships between nearby samples and color channels to estimate a complete image, but fine patterns can still create aliasing or false colors.

The Bayer pattern is one sensor design, not the only possible color-capture arrangement. Alternative mosaics, stacked sensor designs and multiple-sensor systems make different trade-offs. A rendered RGB file usually no longer contains the original mosaic in its directly interpretable raw form.

Three practical examples

A raw converter reads the correct Bayer ordering before reconstructing a camera image.
A camera engineer evaluates how fine fabric patterns interact with the filter array and demosaicking method.
A vision team decides whether to process sensor mosaics directly or use the camera's rendered RGB output for a particular task.

Think of it this way

Think of covering a grid of light meters with different colored windows, so each meter observes only the light passed by its own window. Neighboring readings must then be combined to estimate the full color at each location. The analogy is limited: real filter sensitivities overlap, sensor responses require calibration, and interpolation cannot recover every detail lost through spatial sampling.

Worked example

Consider a four-by-four array formed by repeating an RGGB unit. It contains sixteen photosites: eight green-filtered, four red-filtered and four blue-filtered. It therefore provides sixteen individual filtered measurements, not forty-eight independently measured RGB channel values.

After demosaicking, an output with sixteen RGB pixels contains forty-eight channel values. Some values correspond to directly sampled components and others are estimated using neighboring information and the reconstruction method.

This does not mean the camera simply triples its measured information. The extra channel values complete the representation through estimation. If the raw mosaic is interpreted with the wrong filter order, red and blue information or green positions may be assigned incorrectly, producing strong color errors and reconstruction artifacts.

Illustrative case study

Fictional example: Sensorbridge Lab connects a new camera module to an inspection prototype. Its first raw preview has an unusual color cast and alternating edge artifacts. Engineers verify the sensor's documented mosaic order and the image orientation before changing the color profile.

They discover that the processing code assumed a different starting filter position. After correcting the interpretation, they compare demosaicking options using fine markings and neutral reference regions. The raw samples remain available for reproducible tests.

This fictional case illustrates a basic integration check. It does not imply that every color cast comes from Bayer ordering; illumination, gains, spectral response and color conversion can also contribute.

Common misconceptions and mistakes

A Bayer photosite is not a complete RGB pixel in the same sense as a rendered image pixel. Counting output channels as independent sensor measurements overstates the captured information.

The mosaic also does not guarantee artifact-free color. Repeating textures can interact with sampling and reconstruction to produce moiré or false color. More green sites do not replace white balance or color calibration. Finally, a camera's megapixel count does not fully describe its color resolution, optics, noise or processing quality.

Business use

Understanding Bayer sampling helps teams interpret camera specifications and diagnose raw-processing problems. It is relevant to product imaging, embedded vision and any workflow where fine detail or color consistency matters.

Record the mosaic order, raw bit depth, black levels and processing conventions when integrating a sensor. Test patterns and representative real scenes can expose reconstruction weaknesses that a general photograph hides. Compare the final task result, such as reading a mark or detecting a defect, rather than assuming that a sharper-looking demosaicked image is necessarily more accurate.

Check your understanding

Does a sixteen-photosite Bayer sample directly measure sixteen complete RGB triplets?

Answer: No. It records one filtered measurement at each site. Demosaicking estimates the missing components to create the complete RGB representation.

Related terms

Color filter array (CFA)
Demosaicking
Aliasing (factorial experiments)
Moiré
Adaptive Homogeneity-Directed (AHD) demosaicking
Colour filter array
photosite
raw image
aliasing
colour calibration
Adaptive Homogeneity-Directed demosaicking

Learn next

Draw a four-by-four Bayer mosaic and count the red, green and blue positions. Mark the missing color components at one red site. Then compare a raw mosaic view with two demosaicked versions of the same capture, paying attention to thin lines and repeated textures rather than only overall color.

Sources

Bryce E. Bayer, Color imaging array, original patent document: https://patents.google.com/patent/US3971065A/en Hirakawa and Parks, Adaptive Homogeneity-Directed Demosaicing Algorithm: https://doi.org/10.1109/TIP.2004.838691 The sample count and camera-integration case are original illustrations. The patent is cited for technical history, not for a current legal-status opinion.

Previous: Baum - Welch algorithm Next: Bayes-adaptive Markov decision process