How reverse image search works

From picture to fingerprint

When you give a search engine an image, it does not compare pixels one by one. Instead it analyses the picture and produces a compact description of its content, often called a fingerprint or an embedding. This description captures things like shapes, edges, colours, textures and the arrangement of objects.

Two photos of the same scene taken a moment apart end up with very similar descriptions, even if they are not identical pixel for pixel. That is why a resized or lightly edited copy can still be recognised.

Building the index

Before you ever search, the engine has already crawled huge numbers of web pages and processed the images on them in the same way. Each image's fingerprint is stored in an index together with the page it was found on and other details such as size and date.

When you search, the engine compares your fingerprint with the stored ones and returns the closest matches. This is why the index matters so much: an image that was never crawled can never be matched.

Exact copies versus similar pictures

Engines can run two kinds of matching. Near-duplicate matching looks for the same image, possibly resized, compressed or lightly edited. Similarity matching looks for pictures that resemble yours in subject or style, such as other photos of the same landmark or products of the same kind.

Some engines lean towards one kind. TinEye is built around finding copies and modified versions of the same picture. Google Lens leans more towards similarity and recognition, which is why it also suggests products and related subjects.

Why results differ between engines

Each engine has its own crawler, its own index, its own matching models and its own ranking rules. They also reach different parts of the web, and their coverage of languages and regions varies. The result is that the same image can return very different pages in different engines, which is why comparing several is worth the effort.

What makes matching harder

• Heavy cropping, so that only a small part of the original remains.
• Mirroring, rotation, strong filters or colour changes.
• Text, stickers or frames added on top of the picture.
• Very low resolution or heavy compression.
• Screenshots that include menus, buttons and borders.
• Images that were never published publicly or that sit behind logins.

AI-generated and edited images

Images created by AI tools usually have no earlier source, because they were generated rather than photographed. A reverse image search that finds nothing is therefore not proof of anything, and a match to an unrelated picture does not prove an image is real or fake either. Edited images can break matching in the opposite way, since heavy edits can make a picture look new to the engine.

What reverse image search cannot tell you

It finds pictures, not truth. It can show where an image appears and when, but it cannot confirm who is in a photo, whether a caption is accurate or whether an image was edited. Use it as one clue among several, and combine it with context, dates and other sources before drawing conclusions.