Understanding the TinEye API score

Each result includes a score from 0 to 100 indicating how much the matching image has been modified from the query image. By default, the results are ordered by descending scores so the first match returned is the strongest match.

Weaker matches may include partial matches and matches that are not relevant to your needs. For example, a logo, some text or the background in the result image may match part of your search image, but the overall images are not the same. This is the nature of our image searching: it looks for identical regions in images and cannot decide that a matching logo, text or background is a bad or irrelevant match for you.

A common mistake is attempting to use a score cut-off to eliminate bad matches. We generally recommend against this practice because it can easily eliminate good matches, too. Here are a few examples where this is the case.

When searching for the image above, you may find matches where the image is cropped or resized. Both of these are valid matches but their scores are only 22.4 and 24.7, respectively.

On the other hand, consider the following matching images of hummingbirds. These are clearly two different images, and yet the match score is 51.4, significantly higher than the legitimate matches shown above.

Some of our clients, especially very high-volume clients, still decide to use a score threshold to eliminate weaker matches. This is fine, as long as your use case does not require finding small or heavily modified versions of images.