Amazon Rekognition is the one of the latest additions to the Amazon Artificial Intelligence (AI) suite of services and draws on two decades of in-house artificial intelligence expertise. Rekognition is a Deep Learning-based image detection and recognition service designed to be incorporated into the larger Amazon Web Services (AWS) ecosystem. When used in conjunction with services such as Amazon Lambda serverless compute and Amazon S3 elastic storage, Rekognition can deliver scalable, reliable, low-cost solutions with a rich set of features.
Introduced in AWS re:Invent 2016 back in November, Rekognition offers five key features for developers:
- Object & Scene Detection –identify objects, scenes, and image concepts
- Facial Analysis – detect a human face and key facial characteristics such as sentiment (“happy, “sad”, etc.), landmarks (“eyes open”) and demographics (male, Asian, etc.).
- Face Comparison – compare against one or multiple faces simultaneously
- Facial Recognition – identify one or multiple faces and display their corresponding attributes
- Age Estimation – estimate age range
When Amazon Rekognition is queried for an image search, it compares the submitted image against its database of billions of images and returns its search results. These results contain attributes of the image known as labels. Labels can include such attributes as “dog”, “beach”, “happy”, “blonde”, etc. Amazon typically returns labels for images with an accuracy of 50% or greater. However, an important feature of Rekognition is the ability for you to adjust this accuracy percentage filter based on your needs. If you were using Rekognition as part of an access control solution, for example, you would want the accuracy of the search results returned to be much closer to 100% if you were using its facial recognition features in determining whether or not to allow access for an individual.
The number of use cases for Amazon Rekognition is quite large. Retailers could use it to measure and improve in-store customer experiences by monitoring shoppers’ sentiments in real-time. Rekognition could be incorporated with Intel’s RealSense cameras for access control solutions for buildings or even your computer. Advertisers could possibly use the service to deliver custom, personalized marketing messages to you via digital billboards. Rekognition also be deployed as part of a mass surveillance solution to identify people or objects of interest or concern, especially in areas where there is many people in close proximity such as a sporting event or tourist location. With an estimated 1.2 trillion consumer images alone expected to be uploaded to Amazon in 2017 and our necessary focus on national security, there is no shortage of problems for Amazon Rekognition to help solve.
To Learn More: