> ## Documentation Index
> Fetch the complete documentation index at: https://docs.burst.trade/llms.txt
> Use this file to discover all available pages before exploring further.

# Visual Clustering

> How Burst groups similar token images using perceptual hashing.

Image resolution is not based on exact byte-for-byte matches. Burst uses visual clustering to handle the reality that the same meme gets uploaded in many slightly different forms.

## How visual clustering works

Burst groups similar images into visual clusters using perceptual hashing.

That allows edited or resized versions of the same meme to resolve together.

It also lets a newer meme variant replace an older one naturally when it gains clear support.

## When images update

An image update requires:

* one visual cluster to become clearly dominant,
* a clear margin over the current image cluster,
* persistence over time,
* and support from multiple distinct deployers.

## Why perceptual hashing

Exact file matching would miss:

* resized versions of the same image,
* images with minor edits or overlays,
* re-exports at different quality levels,
* and cropped or padded variants.

Perceptual hashing compares the visual content of images rather than their raw bytes. This makes clustering robust to common image variations.

## Anti-spam protection

The multi-deployer requirement prevents a single actor from flooding an identity cluster with a new image to force an override.

Combined with the dominance and persistence requirements, this makes image manipulation through spam significantly harder.