Is face recognition correct?

The facts about facial recognition with artificial intelligence

A similarity score is a statistical measure of how likely two faces are the same person in an image analyzed by Amazon Rekognition. For example, an image with a similarity rating of 95% would mean that among all the faces analyzed by Rekognition, this image has a 95% similarity to the face that is the subject of the search. The higher the similarity score, the more likely the two images contain the same person. However, even 99% similarity does not guarantee that it is a match.

Because Rekognition uses a so-called probabilistic system, in which results cannot be determined with absolute accuracy. Instead, it is a prediction.

This is where the similarity threshold comes into play. A similarity threshold is the lowest similarity score that the application will accept as a possible match using Rekognition. The selection of the threshold has a significant impact on the search results returned. The number of misidentifications (sometimes called "false positives") that the customer receives is a direct result of the threshold setting. A customer selects the appropriate setting based on their requirements and the area of ‚Äč‚Äčapplication of the application.

We recommend a threshold setting of 99% for scenarios where extremely close matches are important. In public safety and law enforcement, for example, this is often an important first step in narrowing the field of possible hits so that people can use their judgment to go through the remaining results more fully.

On the other hand, many Amazon Rekognition use cases do not require human verification. For example, the secondary factor authentication with an employee ID and a face recognized by Amazon Rekognition with a high (99%) similarity threshold. For a personal photo collection application that can tolerate a few false hits, a lower threshold of 80% may be acceptable. Customers can adjust the similarity threshold according to their usage scenario and requirements.