Amazon Rekognition Announces Real-Time Face Recognition, Support for Recognition of Text in Image, and Improved Face Detection

Amazon Rekognition today announces three new features: detection and recognition of text in images, real-time face recognition across tens of millions of faces, and detection of up to 100 faces in challenging crowded photos. Customers who are already using Amazon Rekognition for face verification and identification will experience up to a 10% accuracy improvement in most cases.

Text in Image

Customers using Amazon Rekognition to detect objects and faces in images have been asking us to recognize text embedded in images. Examples of this text include street signs, license plates captured by traffic cameras, news, captions on TV screens, and stylized quotes overlaid on phone-captured family pictures. Starting today, you can use Rekognition Text in Image to recognize and extract textual content from images. Text in Image is specifically built to work with real-world images rather than document images. It supports text in most Latin scripts and numbers embedded in a large variety of layouts, fonts, and styles. It also supports recognition of text overlaid on background objects at various orientations, such as banners and posters.

“As a visually-driven platform, Pinterest relies heavily on the speed and quality of images, but the text behind those images is just as important, as it provides context and makes Pins actionable for our 200M+ active Pinners. In working with Amazon Rekognition Text in Image, we can better extract the rich text captured in images at scale and with low latency for the millions of Pins stored in Amazon S3. We look forward to continuing to develop the partnership with AWS for high quality and fast experiences for Pinners and businesses on Pinterest.” – Vanja Josifovski, CTO, Pinterest

“Professional photographers often use SmugMug to share and sell photos containing text, for example, the numbers on marathon race bibs. Amazon Rekognition Text in Image allows us to programmatically extract bib numbers at scale and provide event photographers with even more functionality to quickly and easily share and monetize photos from these events. ” – Don MacAskill, Co-founder, CEO & Chief Geek at SmugMug

Real-time Face Recognition

You can now perform real-time face searches against collections with tens of millions of faces. This represents a 5-10X reduction in search latency, while simultaneously allowing for collections that can store 10-20X more faces than before. For security and public safety applications, this update helps identify persons of interest against a collection of millions of faces in real-time, enabling use cases that require an immediate response.

Crowd-Mode Face Detection

Starting today, customers can also detect, analyze, and index up to 100 faces (up from 15) in a single image. With this improvement, you can accurately capture demographics and analyze sentiments for all faces in group photos, crowded events, and public places such as airports and department stores.

“We have large collections of photos that our users bought or uploaded to our platform in the past. We often need to search for photos of a particular user’s child within these collections, and have been using Amazon Rekognition for this. As we have lot of group photos with dozens of small faces, we previously had to crop and divide the original image to detect all faces correctly. By using the new crowd face detection feature, now we can easily detect all the faces in one go without any complicated preprocessing.” – Shinji Miyazato, Engineering Department SRE Lead, Sen Corporation.

Improved Accuracy for Face Detection Model

We have also improved the accuracy of our face detection algorithms, which provide up to a 10% improvement in the accuracy of face verification and identification applications used at check-in counters and employee turnstiles, and used in mobile face-based authentication.

What excites us most is the production rollout of large-scale image analysis workloads by customers from multiple verticals. Some noteworthy new additions to the growing list of customers include the following:

Butterfleye, a security camera provider for the home and small business market used Amazon Rekognition to develop quickly and cost-effectively their facial and object detection camera, reducing their development time from 18 months down to 4 months and saving over $1M in R&D expenses.

Open Influence helps Enterprise customers find social media influencers to promote their brands. By being able to easily integrate Amazon Rekognition with their own data pipeline, they can generate quality search returns that help their customers uncover influencers they would have never otherwise found.

Amazon Rekognition is helping Marinus Analytics fight human trafficking. Their flagship software is used by US law enforcement agencies on sex trafficking investigations. With Amazon Rekognition, investigators using Marinus Analytics are able to take quick and effective action by searching through millions of records in seconds to find victims.

Find out more about Amazon Rekognition. You can also get started quickly with Amazon Rekognition. If you have any questions, please leave them in the comments.

Ranju has been with Amazon for almost five years and leads Amazon Rekognition, a deep learning-based image recognition service which allows you to search, verify and organize millions of images. Before joining Amazon, Ranju worked at Barnes and Noble leading Nook Cloud engineering. His team was responsible for strategy, design, development and SaaS operation of Nook mobile services and Digital Asset Management Services.