Amazon Rekognition

Amazon Rekognition labels a woman's face including sentiment analysis

Amazon Rekognition

Amazon Rekognition (sometimes AWS Rekognition) is a pre-trained computer vision AWS deep learning algorithm which can pull and index insights from images and videos. While Amazon Rekognition is fully customizable, out-of-the-box it provides the following services:

  • automated content moderation
  • facial recognition
  • sentiment analysis
  • labeling/metadata
  • text detection
  • celebrity recognition
  • video segment detection

Amazon Rekognition Features

Automated content moderation

Amazon Rekognition can detect things like pornography, violence, foul language, drug or alcohol use, etc. to automate the process of content moderation. Not only is this faster and cheaper than using a human, but can prevent people from needing to look at potentially offensive images. More about automated content moderation.

Facial recognition

Amazon Rekognition can detect faces appearing in images and videos and recognize attributes such age range, eyeglasses, gender, and more. This is part of the face detail feature which includes emotions or sentiment analysis.

Sentiment analysis

Beyond simply recognizing faces in images or video, Rekognition can recognize if are the people are feeling one of the following emotions, along with the confidence-level it has:

  • angry
  • calm
  • confused
  • disgusted
  • fear
  • happy
  • sad
  • surprised

Labeling/metadata

One of the most powerful features of Rekognition —which is also customizable— the machine can label images with things it observes like:

  • Activities (running, playing, laughing, etc.)
  • Objects (trees, streets, cars, people, etc.)
  • Scenes (beach, sunset, city, etc.)
  • Concepts (outdoors, fun, etc.)

Text detection

Images and videos often contain text (billboards in the background, t-shirts, etc) and Amazon Rekognition knows how to read. The text along with confidence level can passed along via the detect text feature.

Celebrity recognition

In addition to identifying faces, the algorithm knows "celebrities" out of the box. Jeff Bezos or Werner Vogels are perfect examples.

Video segment detection

A valuable tool for people operating in the media & entertainment industry, Rekognition can automatically identify (and timecode) things like:

  • Black frames
  • Color bars
  • Content (regular video)
  • End credits
  • Opening credits
  • Slates
  • Studio logos

Amazon Rekognition use cases

  • Automated Content Moderation - use machine learning to identify inappropriate content across image and video based on general or business-specific standards and practices.
  • Identity verification - use facial recognition in your user authentication to remotely verify identity.
  • Smart home alerts & automation - Delivering alerts when something is detected in your home live video streams as well as automation like automatically turning on the light when a person is detected
  • Video key frame detection - Automatically detect key video segments to reduce the time, effort, and costs of video ad insertion, content operations, and content production.

Rekognition pricing

Amazon Rekognition pricing operates on a pay-as-you-go model. You pay only for what you use. All of our pricing calculations were done in 2022, using the US West Oregon region, which tends to be both price competitive and one of the earlier regions features are launched. 

Image processing cost

  • image analysis generally ranges between $0.001 per image (for the first million images) up to $0.00025 per image if you are processing 35 million images or more.
  • label detection starts at $0.0010 per image, aka.1,000 images for 1 dollar.

Video processing cost

Video analysis is broken down into stored video analysis and streamed video analysis. Both types of video analysis are priced on per minute per service basis, with all of the following services billed at $0.10/min for any stored video:

  • label detection
  • content moderation
  • text detection
  • face detection
  • celebrity recognition
  • face search
  • person pathing

Shot detection and Technical Cues Detection (silent black frames, credits, etc) are billed at half the rate for other services ($0.05/min).

When it comes to streamed video, currently Label Detection and Face Search are the only available services, with Face Search billed slightly higher ($0.12/min) vs. stored video, and Label Detection strangely billed at slightly less ($0.00817/min).

Rekognition custom labels pricing

When it comes to custom labels —one of the most powerful features of Amazon Rekognition— pricing gets tricky, as it is based on training and inference (or usage) hours. The good news is that you can get started with Amazon Rekognition Custom Labels for free. The Free Tier lasts 3 months and includes 10 free training hours per month and 4 free inference hours per month. Once you surpass that level pricing they charge as follows:

  • Training - $1.00/hour
  • Inference (usage) - $4.00/hour

If you are considering an Amazon Rekognition project that requires custom labels, it's a good idea to get help (we'd be happy to chat). The platform is incredibly valuable, but there is a real danger of unintentionally over running on costs especially when just getting started.

More robust pricing can be found on the Amazon Rekognition pricing page.

Customizing Amazon Rekognition

Out-of-the-box functionality in AWS deep learning algorithms is great, but it can also be extended. Most of this functionality boils down to custom labeling but this process is exceedingly powerful. Here are a few ideas:

  • It could be trained to recognize specific people, not just celebrities.
  • It could be trained to recognize clothing styles.
  • It could learn when pets are misbehaving.

The possibilities are almost endless, and we are just starting to see more and more creative application. As with any algorithm, training can be complicated but if you have an idea we'd be happy to talk with you about it.

    Date posted: February 1, 2022

    Add new comment

    Restricted HTML

    • Allowed HTML tags: <a href hreflang> <em> <strong> <cite> <blockquote cite> <code> <ul type> <ol start type> <li> <dl> <dt> <dd> <h2 id> <h3 id> <h4 id> <h5 id> <h6 id>
    • You can enable syntax highlighting of source code with the following tags: <code>, <blockcode>, <cpp>, <java>, <php>. The supported tag styles are: <foo>, [foo].
    • Web page addresses and email addresses turn into links automatically.
    • Lines and paragraphs break automatically.

    Learn more about how we've set our clients up for success by reading our case studies

    About the Author

    Joaquin Lippincott, CEO

    Joaquin is a 20+ year technology veteran helping to lead businesses in the move to the Cloud. He frequently speaks on panels about the future of tech ranging from IoT and Machine Learning to the latest innovation in the entertainment industry.  He has helped to modernize software for industry leaders like Sony, Daimler, Intel, the Golden Globes, Siemens Wind Power, ABC, NBC, DC Comics, Warner Brothers & the Linux Foundation.

    As the CEO and Founder of Metal Toad, an AWS Advanced Consulting Partner, his primary job is to "get the right people in the room".  This one responsibility is cross-functional and includes both external business development functions as well as internal delegation and leadership development.

    A UCLA alumni, he also serves in the community as a Board Member for the Los Angeles Area Chamber of Commerce, the Beverly Hills Chamber of Commerce, and Stand for Children Oregon - a public education political advocacy group. As an outspoken advocate for entry-level job creation in tech he helped found the non-profit, P4TH, an organization dedicated to increasing the number of entry-level jobs in the tech industry, and is in the process of organizing an Advisory Board for the Bixel Exchange, a Los Angeles non-profit that provides almost 200 tech internships every year.

     

    Have questions?