How to georeference an image
Georeferencing ties an image to real-world coordinates. After it, a map program knows where each pixel is on the ground.
What you need
- The image, as JPEG, PNG or TIFF.
- A rough idea of where it was taken.
- Something to compare it with: a current satellite image, a map, or an already-placed image.
The idea
You find points that appear in both your image and the reference, such as road junctions, field corners or building edges. Each pair says that this pixel in your image is at this place on the ground. With enough pairs, a transform can be fitted that places the whole image.
Doing this by hand means choosing the pairs yourself, usually at least four to ten of them, spread across the image. It works, but it is slow, and it gets tedious across many images.
The automatic way
- Open the tool and load your image.
- Tell it where the image is: search for a place, draw a box, or type coordinates.
- Choose a reference. Sentinel-2 works worldwide. NAIP is sharper in the United States for small areas.
- Press Match. It finds and pairs features in both images, and reports how many pairs agreed and how well.
- Export the world file, or the points for QGIS if you want to check or adjust the fit.
Check the result
Look at the numbers it reports: matches, inliers and residual. A few dozen consistent matches and a small residual is a good sign. A handful of matches means the transform is a guess. Then open the image in QGIS over a basemap and check that roads and coastlines line up across the whole frame, not only the middle.
When it does not work
Open water, desert, forest and snow offer little to match. Very old images of areas that have been rebuilt may share too little with today's imagery. A smaller search area often helps. See the FAQ for more.