Reverse image search

How does it work?

The engine reduces the uploaded picture to a compact description of its visual content, historically a perceptual hash and today usually a learned feature vector, then looks for pages whose images have a close description. Because the comparison is of whole images, the strongest matches are exact copies, followed by crops, resized versions, recolourings and images with small overlays.

What does it find and not find?

How is it different from a reverse face search?

A reverse face search compares faces, not pictures, so it can find a different photo of the same person. A reverse image search can only find versions of the picture it was given. For a stolen-photo problem both are useful: the image search finds where the exact picture was reposted, the face search finds other pictures of the same face. The product site's guide to Google Lens covers the overlap in practice.

Which engines offer it?

• Google Lens, inside the Google app, Chrome and images.google.com.
• TinEye, the oldest dedicated engine, strong on finding the first appearance of an image.
• Yandex Images, often the strongest free option for similar faces, still image-based.
• Bing Visual Search, with product and object matching.

Where does it go wrong?

Mirrored, heavily cropped or filtered copies can slip through, newly generated images have no copies to find, and pages behind a login are never indexed. Results also depend on the engine's crawler having visited the page; a copy on a small forum may be absent from every engine.