Skip to content

A graph-based tool helps researchers match wildlife photos to individual animals

Research · Other research

Photographs can help researchers follow individual wild animals without capturing or marking them. But turning thousands of images into reliable records of which photos show the same animal can take substantial human effort. A new system called β-Stability is designed to reduce that workload while keeping people involved in important decisions. The findings are reported in a bioRxiv preprint that has not yet been peer reviewed.

The researchers evaluated it on seven image datasets, including photographs of Grevy’s zebras, giraffes, lions (Panthera leo), belugas, forest elephants and whale sharks.

Connecting photographs into a network

β-Stability represents each photograph as a point in a network and each possible match between two photos as a link. The links record whether the photos are thought to show the same or different animals, along with how confident that match is. The system adjusts the links automatically, then directs human reviewers to the decisions most likely to improve the consistency of the overall grouping. Photos joined into the same group are assigned to one individual.

Across the seven datasets, the method produced more accurate identity assignments than previous approaches for the same amount of review. In a dataset of 3,764 Grevy’s zebra photographs, it reached the correct population count after 580 human reviews. The CEAF F1 score was about 0.99, indicating that the photo groups closely matched the known identities; the correct count did not mean every assignment was perfect.

The system also remained the best-performing method across all seven datasets when 5% errors were introduced into expert reviews. Its final F1 score fell by an average of only 0.041. This suggests the approach can tolerate some mistakes in human review while still producing reliable groupings.

Finding errors in existing records

β-Stability was also tested as an auditing tool on a dataset that had already been curated. It identified 60 labeling errors, and the reported F1 score rose from 0.988 to 0.996 after those errors were corrected.

Reliable identification of individuals in photographs can support studies of animal abundance, movement, behavior and population change. β-Stability does not itself estimate those ecological trends; it aims to make the identity records that researchers rely on more dependable, both when building them and when checking existing data.

Sources and image credit

  1. β-Stability: Efficient, human-in-the-loop, graph-based curation of individual animal identities from imagesPrimary preprint record. The page and full-text PDF were inaccessible during verification; the drafts are based on claims available in the abstract. This is a preprint, not a peer-reviewed journal article.
  2. β-Stability: Efficient, human-in-the-loop, graph-based curation of individual animal identities from imagesSecondary record reproducing the abstract; used to check the study scope, methods and reported results when the primary full text was inaccessible.

Image: A lion (Panthera leo) rests on a tree trunk in Tarangire National Park, Tanzania, on 24 May 2024. Photo: Diego Delso / Wikimedia Commons, CC BY-SA 4.0. Representative photograph. · CC BY-SA 4.0 · Source

About this content: This story was produced with AI assistance within an editorial workflow developed by Wildlife Vagabond. Editorial responsibility remains with Wildlife Vagabond.How AI is used

Other research

SPECIES IN THIS STORY

Species in this story

Lion Panthera leo Explore species

Independent research and conservation news archive

Did you find this information useful?

Wildlife Vagabond is independently built and maintained. Voluntary support helps cover source verification, hosting and continued work on research and conservation news.

Support Wildlife Vagabond

The news archive will remain freely available.