01What it does

Location from the image itself.

Screenshots, social-media downloads, and re-encoded images often arrive without location metadata. Find Region works from the visible image instead.

Find Region is trained on a large, geographically diverse collection of geotagged images. During training, it learns relationships between visual patterns and where images were captured. When a new image is submitted, Raven uses those learned relationships to rank candidate locations.

Each candidate includes a place, coordinates, an approximate search radius, and a similarity score. The results are investigative leads, not confirmed locations.

02Worked example

A balcony, a coffee cup, and out-of-focus trees.

This frame contains no landmark, sign, readable text, or location metadata. Raven returned Austin, Texas as the top candidate, with four of five results concentrated in the same area.

Low-context phone photograph of a coffee cup on a balcony with out-of-focus trees in the background
Input: Single frame · no landmarks · no EXIF or GPS
Raven Find Region result showing Austin, Texas, United States at 90 percent similarity with 4 of 5 results in this region
Result: Austin, Texas · 30.304595, −97.726650 · approximate 15-mile radius · 90% similarity · four of five candidates in the same area
What an investigator can observe
Visible cues
1Light and shadows 2Vegetation 3Objects and materials 4Foreground surfaces
Each detail can narrow the possibilities. None identifies the location on its own.
How Find Region works
Learned geographic patterns
Environment Structures Infrastructure Spatial relationships
Trained on geotagged imagery, Raven learns how combinations of visual patterns relate to geography. It uses those learned relationships across the full frame to rank candidate regions.
No single clue has to identify the place. Raven considers the scene together and returns locations for an investigator to review.
03What you get back

Ranked candidates for investigator review.

  • Place: city, state or region, and country.
  • Coordinates: a latitude and longitude point.
  • Radius: the approximate area around the candidate.
  • Similarity score: a ranking signal used to order the results.

Raven also shows whether several candidates cluster in the same area. In this example, four of five results fell near Austin. The fifth result remained visible as an outlier.

Similarity scores rank candidates against one another. They are not probabilities that a location is correct.

Ranked Find Region candidates from the worked example above: Austin at 90% similarity, Austin at 89%, Austin at 88%, Lakeway at 80%, and Stockholm, Sweden. Four of the five fall in the same area.
01
Austin
Texas, United States
90%
02
Austin
Texas, United States
89%
03
Austin
Texas, United States
88%
04
Lakeway
Texas, United States
80%
05
Stockholm
Stockholm, Sweden
—
04Where it fits

Start broad. Refine when needed.

Find Region narrows an unknown image to candidate areas. When street-level coverage is available, investigators can pass the same frame and selected region to Find Street for further refinement.

Note. Find Region returns investigative leads, not confirmed locations. Similarity scores rank candidates and should not be interpreted as probabilities. Results should be corroborated through normal investigative procedures before action is taken.
05Questions

Common questions.

Does Find Region need EXIF or GPS metadata?

No. Find Region works from the visible image, so screenshots, social-media downloads, and re-encoded files can still be processed.

How does Find Region work?

Find Region is trained on geotagged imagery and learns relationships between visual patterns and geography. It uses those learned relationships to rank candidate locations for a new image.

What does the similarity score mean?

It ranks candidates by visual similarity. It is not a probability that the location is correct.

How precise is the result?

Find Region returns candidate areas with an approximate radius. Use Find Street for street-level refinement or Find Property to match a specific interior or exterior within a supported city.

What kinds of images work best?

Outdoor images with meaningful environmental or structural context generally provide more information. Recognizable landmarks can help, but they are not required.

Who can use it?

Raven is available to verified agencies and investigative teams. Access is granted after review.

New to the discipline? Our guide to image geolocation covers what it is, why metadata so often falls short, and how investigators locate photos without it.