Is zippering the same as moire?

No. Both can involve sampling and reconstruction, but moire usually refers to larger interference-like patterns, while zippering describes alternating teeth along an edge.

Does it only affect color?

No. The pattern can appear as light - dark alternation, although false color may accompany it.

Can a different demosaicking method help?

Yes, depending on the scene and source data. Improvement should be checked for lost detail and other artifacts rather than judged only by smoothness.

AI Glossary / Computer vision and multimodal AI / Intermediate

Zippering

Definition

Zippering is an image artifact in which alternating light, dark or colored pixels create a tooth-like pattern along an edge. In digital photography, it is commonly associated with errors in demosaicking: reconstructing full color from a sensor that samples different color channels at different locations.

What it means

A sharp boundary should often appear continuous, but reconstruction can make adjacent edge pixels fluctuate in a regular pattern. At sufficient magnification, the result resembles the interlocking teeth of a zip.

The term describes the visible artifact, not a physical texture that must have existed in the scene. It is important to distinguish it from a genuine repeating pattern, ordinary stair-step aliasing, sharpening halos and color fringes, although several artifacts can occur together.

Zippering also is not related to ZIP file compression. A raw-processing error can remain visible in a losslessly stored image. Changing the final file format does not necessarily address the stage that introduced it.

How it works

In a typical Bayer color-filter array, each sensor location measures only one color component. Demosaicking estimates the missing components from nearby measurements.

Near an edge, some neighbors can belong to the bright side and others to the dark side. Interpolation that mixes across the boundary can pull an estimated value away from the appropriate side. Because the measured and missing components alternate across the sensor pattern, the errors can form an alternating edge structure.

Edge-aware methods try to use information along a boundary rather than across it. Other methods exploit relationships between color channels. These approaches reduce some failures but still face ambiguity in fine textures, noisy regions and complex intersections.

Later sharpening or resizing can accentuate or obscure the pattern. Diagnosing its origin requires comparing processing stages, rather than assuming the final visible artifact identifies one cause with certainty.

Three practical examples

A raw-converter evaluation inspects high-contrast diagonal and near-vertical edges for alternating reconstruction errors.
A product-imaging team checks whether apparent serrations on a photographed component remain when the same raw capture is processed differently.
A camera-quality experiment compares demosaicking methods on a synthetic mosaic with known full-color reference data.

Think of it this way

Think of repairing a black-and-white tiled border by guessing missing tiles from their neighbors. If some guesses mix tiles from opposite sides, alternating repaired positions can look unlike the correctly measured ones.

The analogy is limited because camera data are intensity measurements, not fixed black or white labels. Real boundaries can be blurred, colored or textured, and the correct missing value is not always obvious from a small neighborhood.

Worked example

Averaging across an edge

Consider a toy green-channel reconstruction beside a vertical boundary. The bright side has value 100 and the dark side value 0. At a missing green sample just inside the bright side, suppose the upper, lower and right neighbors are 100, while the left neighbor is 0.

A four-neighbor average estimates (100 + 100 + 100 + 0) / 4 = 75. The intended bright-side value in this idealized scene is 100. If neighboring edge positions contain measured green values of 100, a row or column of such positions can alternate between measured 100 and estimated 75.

An edge-directed estimate using only the upper and lower neighbors gives (100 + 100) / 2 = 100. In this constructed example, it avoids mixing across the boundary.

Now introduce a thin real feature or noise that changes those vertical neighbors. The simple directional rule may choose poorly or miss genuine structure. The arithmetic demonstrates a failure mechanism and a possible remedy, not a universal demosaicking algorithm.

Testing against a known reference is what distinguishes reduced reconstruction error from merely making the edge smoother.

Illustrative case study

Illustrative case study - fictional

A fictional online hardware seller notices a repeating tooth-like pattern along the edge of a small metal bracket in catalog photographs. The item itself has a smooth edge, but the artifact becomes prominent after aggressive sharpening.

The imaging team compares a lightly processed raw conversion with the final export and tries another demosaicking setting. It also photographs the edge at a slightly different orientation to see how the pattern relates to sampling. The original measurements and controlled comparison point toward processing rather than a manufacturing defect.

The team selects a conversion that preserves useful edge detail without amplifying the alternating pattern. This illustrates diagnosis, not a measured quality improvement or a guarantee that all edge artifacts share the same origin.

Common misconceptions and mistakes

Zippering is not evidence that the photographed object has a physically serrated edge.
Saving an image without lossy compression does not remove artifacts introduced during reconstruction.
Stronger sharpening can make alternating errors more obvious instead of restoring true detail.
Smoothing can hide the pattern while also erasing genuine texture or narrow features.
Edge-aware interpolation helps in suitable cases but cannot resolve every ambiguity caused by limited sampling.

Business use

The business relevance is image credibility and measurement reliability. A processing artifact can be mistaken for a product flaw, alter the appearance of fine detail or interfere with an inspection system.

When evaluating cameras or raw converters, include sharp boundaries at several orientations and fine repeating patterns. Compare crops at a consistent scale and review the intended display size as well. Keep sharpening and resizing settings controlled. If an edge drives a quality decision, confirm the suspected feature through another capture or independent measurement rather than relying solely on an enlarged processed image.

Check your understanding

A processed edge alternates between values 100 and 75, but a known reference edge is uniformly 100 on that side. Does sharpening the alternation establish additional real detail?

Answer: No. It can amplify reconstruction error rather than recover genuine structure.

Related terms

Learn next

Reconstruct the toy missing sample using four-neighbor and along-edge averages. Then change the edge orientation or introduce a one-pixel feature. Identify where each assumption breaks and why a known reference is useful for judging the result.

Sources

Sources: MIT computational-photography material, Demosaicking: https://people.csail.mit.edu/fredo/comp-photo-book/03-basic-image-processing-and-isp-12-demosaicking.html ; associated denoising and demosaicking problem set: https://people.csail.mit.edu/fredo/comp-photo-book/22-appendices-05-problem-set-3-denoising-and-demosaicking.html . Relevant Bayer reconstruction, zippering and edge-directed interpolation explanations were reviewed. The numerical edge example is an original illustration, not an executed implementation test.

Previous: Zipf's law Next: Zonal polynomial