Skip to content


Meet AWS Rekognition: Image Analysis and Facial Recognition

September 7, 2017
2 min read

Every day, more businesses realize the value of facial recognition. But the technology is complex, to say the least — which means that incorporating it into any platform or product takes time, even for experienced developers.

Enter Amazon. Last year, AWS launched its image recognition software, Amazon Rekognition. With Rekognition, users can easily add image analysis to their applications and incorporate deep machine learning into whatever they’re building.

How it Works

With a 12-month free account, developers can start creating their own facial recognition systems. Rekognition software delivers quite a few features, including:

  • Object and scene detection: When you upload a photo of an object or scene, Rekognition can identify what they are and return labels, along with a confidence score.
  • Facial analysis: When you upload a photo of a person, a green bounding box identifies the face and returns demographic data, also with a confidence score. That data may include age range, sentiment (happy, calm, angry), or physical appearance (glasses, moustache, etc.).
  • Facial comparison: Rekognition compares images of two faces and returns a similarity score.

With facial comparison, Rekognition works like any computer vision system. It compares a photo to an image database and identifies a match. That requires two primary functions: indexing faces and searching faces.

When Rekognition indexes faces, it takes all your images and turns them into a collection, assigning a face ID to each image. When it’s time to make a match, Rekognition will take a live capture, turn it into a face ID, perform a search and return a similarity score.

Our Experience With Rekognition

Recently, a client asked us to explore the development of a facial recognition engine on a very short timeline. Rekognition seemed like a good fit: it promised scalability, functionality, and speed — which meant we could save countless data science and software engineering hours.

As we began investigating, we found that off-the-shelf Rekognition had some limitations. But using simple statistics, we were able to manipulate the platform and improve the performance. In only three days, we delivered a highly effective solution for the client.

Rekognition isn’t the only image analysis API around. Google, Microsoft, IBM — to name a few — also have developed platforms with similar or overlapping features. Of course, each API has its own strengths (and weaknesses, too). Each can be used to fast-track machine learning, or at least get it off the ground, so we think they’re valuable tools — especially in cases where time and resources are limited.

Ready to get started with Rekognition? Reach out to the team to discuss your next IoT project.