Amazon Rekognition

Amazon Rekognition

Computer vision (CV) capabilities provided by Amazon Rekognition are pre-trained and programmable, making it easy to extract data and insights from images and videos. Pre-trained or configurable computer vision APIs can be swiftly integrated into your apps without the need to build machine learning (ML) models and infrastructure from scratch. Images and videos are analyzed in minutes using artificial intelligence to assist people in visual assessment tasks that would otherwise be performed manually (AI).

Amazon Rekognition Video

Amazon Objects, scenes, celebrities, text, activities, and incorrect material are all analyzed by Rekognition Video on videos saved in Amazon S3. Faces can also be detected, analyzed, and compared in videos using Rekognition Video’s facial recognition technology.

Timestamped results allow you to rapidly build an index for a comprehensive video search or skip to an interesting part of the film for additional examination.

With Rekognition Video, you’ll be able to get the exact coordinates of every object or face you see on the screen via bounding box coordinates. The Amazon Rekognition Video service can also keep an eye on a video feed from Amazon Kinesis Video Streams in order to identify and search for faces. Closed captioning, profanity filtering and video streaming transcription are all possible with Amazon Transcribe and Amazon Rekognition Video.

Use Cases

  • Searching and indexing media assets  

With Amazon Rekognition Video, you can automatically index and search massive video archives by using object, scene, activity, celebrity, text, and face analysis metadata. You don’t have to go through all the videos one by one, which saves you time. These serverless services like media2cloud and media insights engine allow for a seamless transition from tape to MAM system for archive curation, filtering, and revenue monetization of archives.

  • Strict adherence to standards  

Adherence to standards ensures that any inappropriate or brand-unsafe content in your video assets may be identified right away. Amazon Rekognition Video’s timestamps should be examined by your human moderators. Using the hierarchy of moderation labels provided, you may also manage international market compliance requirements. Using Amazon Transcribe metadata, you may control the volume of the audio.

  • Ads in context 

So you may show ads that are most relevant to the video content. Improve the ad’s efficiency and return on investment by doing so.

  • Responsiveness to threats to the public  

With Amazon Rekognition Video, you can create apps that assist you to locate persons who have gone missing in video footage. A missing persons database can be used to quickly find probable matches.


  • Detection of objects, scenes, and activities

Rekognition Video can identify hundreds of items, scenarios, and behaviors, such as delivering a package or dancing, in seconds. You are given a confidence score for each label you recognize. For certain objects like “Person” and “Car,” bounding boxes are also provided, allowing easy counting and localization. A video’s motion can be used by Amazon Rekognition Video to assist with more sophisticated tasks like “blowing out a candle” or “extinguishing a fire”. Using the rich media assets’ extensive metadata, you may improve your content’s searchability and provide advertisements that are relevant to the material that comes before them.

  • Moderation of online content

Each detection by Amazon Rekognition Video comes with a timestamp and a brief description of the content that was detected. For each level of potentially dangerous content, you’ll see a list of classifications and confidence scores. A subcategory of “Explicit Nudity” may be “Graphic Female Nudity,” for example. Using confidence scores and comprehensive labeling, alternative business rules can be established for different markets and areas.

  • Detecting text

Amazon Rekognition Video also provides a location bounding box and a time stamp for each detected text. Filter terms based on ROI, the size of the bounding box, and the level of confidence. As an example, you may only wish to see the bottom third of a soccer game’s scoreboard.

  • Celebrity appreciation

Amazon Rekognition Video makes it simple to recognize well-known actors and actresses in videos. Links to relevant resources, including the celebrity’s IMDb page are included in each celebrity’s name.

  • Recognition and analysis of the face

Using Amazon Rekognition Video, up to 100 faces can be identified in a single video frame. To go along with the timestamps associated with every face that has been detected thus far, researchers have been able to extract information about the person’s gender, emotions, and age.

  • Face-to-face search

An Amazon Rekognition Video search against a private face photo database may recognize known people in the video being viewed. An individual match is given a similarity score, and timestamps are established for each time the same person is spotted in the movie throughout. Each time an unfamiliar person appears in a video, the Amazon Rekognition Video service can deliver timestamps with unique identifiers for each person who appears in the video.

  • Pathing of individuals

People in your video can be tracked by Amazon Rekognition Video, which can track their location and movement. Using Amazon Rekognition, you can keep track of how many people are in a video by generating a unique index for each person it recognizes.

  • Video analysis of a live webcast

The Amazon Rekognition Video service can be used to recognize and search for persons in live video broadcasts. Rekognition Video, for example, can do low-latency face searches against a collection of your images using Amazon Kinesis Video Streams as an input stream.

Amazon Rekognition Image

Image processing software In order to recognize objects, scenes, and faces in images, Rekognition Image employs deep learning. In addition, it detects celebrities, extracts text from photographs, and identifies problematic information in implementations. Also, it lets you search for and compare other people’s faces. Rekognition Image uses the same proven, highly scalable, and deep learning technology that powers Amazon’s other products to analyze billions of photos every day for Prime Photos. In order to make an informed selection about how you intend to utilize the information you have acquired, it provides a confidence interval for everything it finds. As a result, the bounding box coordinates of all detected faces can be utilized to determine the face’s placement in the image, as well as the image’s origin.


  • Identification of Objects and Scene

Rekognition Image can accurately identify tens of thousands of items, including autos, dogs, and furniture. Recognizing items within an image, such as a sunset or a beach, is also possible. Thus, you can browse through large collections of images with ease while filtering and curating them at the same time.

  • Recognition of Facial aspects

Rekognition Image can help you find people who look like you in a large collection of photos. A picture index of all the faces that were found can then be created using your photos. Rekognition Image provides fast and accurate search results since it generates faces that are the most similar to your reference face.

  • FaceTime analysis

Using Rekognition Image, you can see whether or not a smile or open eyes are present on a person’s face in an image. During picture processing, Rekognition Image will deliver the face’s position and a rectangular frame.

  • Face Comparison

An image recognition tool called Rekognition Image can tell you whether two faces in two photos are likely to be the same person. Rekognition uses similarity scores to compare a user to a reference photo in real-time.

  • Detection of insufficient content

In order to better serve your needs, recognition images can recognize explicit and suggestive information. As a result, Rekognition provides a hierarchical labeling list with confidence scores to allow precise control over what photographs you accept.

  • Celebrity Appreciation

An image recognition system called Rekognition Image can identify and locate thousands of prominent figures in their particular fields of activity. In digital picture libraries, this feature provides an index and search capability for photos of famous people.

  • Image with text

Road signs, license plates, t-shirts, and coffee cups may all be scanned with Rekognition Image to find and extract text. Each word or line that Text in Image identifies in the image has its own rectangle frame and confidence score.

  • The detection of Personal Protective Equipment (PPE)

You can tell if Amazon Rekognition Image uses personal protective equipment (PPE) by looking at the image (PPE)

  • Command line, API, or console-based management

As previously mentioned, there are three ways to get to Amazon Rekognition: via API, console, and command-line (CLI). The Rekognition APIs can be accessed using the console, API, and CLI to search a face, identify labels, and analyze faces. AWS Lambda blueprints for Rekognition are also available for Amazon S3 and Amazon DynamoDB, which make it easy to start imaging techniques using events in your AWS Cloud computing data stores.

  • Legal Protection in the Office

There is a connection between Amazon Rekognition and IAM services (IAM). Manage your account’s resources and Amazon Rekognition API access via IAM policies.

  • Human Consideration

With Amazon Augmented AI, you can easily add a human evaluation for risky image identification to Amazon Rekognition (Amazon A2I). Rekognition predictions can be evaluated and validated using Amazon A2I’s built-in picture moderation mechanism. You can utilize your own team of reviewers, or you can use Amazon Mechanical Turk and Amazon A2I to access a workforce of over 500,000 independent contractors currently working in machine learning. Pre-screening workforce vendors for quality and security is also done by AWS computing. In the Amazon A2I development guide, the Amazon A2I Connectivity with Amazon Rekognition section, you may learn more about developing human review workflows using Amazon Rekognition.

Amazon Rekognition Custom Labels

Amazon Rekognition Custom Labels make it possible to identify specific objects and settings in pictures with Amazon Rekognition. Sorting machine parts in an assembly line, identifying healthy and ill plants, or finding animated characters in films are just a few examples of how you may use these techniques to uncover references to your brand on social media.

It can take months to build a custom image analysis model, which requires a lot of work, knowledge, and resources. Additionally, the model needs hundreds or even thousands of hand-labeled photographs in order to draw meaningful results.

In order to categorise all of this data, it takes a significant amount of time and resources over the course of several months.

When it comes to Amazon Rekognition Custom Labels, we take care of all the grunt work. Expanding Rekognition’s capabilities, this add-on uses millions of images from diverse categories to train Rekognition’s algorithms. Take advantage of our simple terminal and upload a few hundred or less training images instead of thousands. Rekognition can get up and running in a matter of seconds with pre-labeled images. If not, Amazon SageMaker Ground Truth can name them for you or you can name them yourself using Amazon SageMaker.

Your picture data can be used to build a custom image analysis model in a matter of hours. A model is trained by loading and evaluating training data, selecting the proper machine learning algorithms, and providing performance metrics.. The Rekognition Custom Labels API can then be used to include your custom model into apps.

Use Cases

  • Measuring brand coverage  

The media coverage of a company’s clients must be accurately reported by marketing agencies. Client brands and products are often tracked manually on social media, television, and sports videos. A model can be trained to recognize a client’s logo and merchandise using Amazon Rekognition Custom Labels. In place of manually monitoring traditional and social media, they can run photographs and video frames through the model.

  • Find syndication content

There are typically thousands of photos and films that must be sifted through by content developers in order to get the right material for their presentations. When it comes to sports, for example, broadcasters have to make their own highlight videos about games, teams, and players. Training custom models to recognize teams, players, and frequent game events like goals scored, suspensions, and injuries can generate a relevant assortment of photos and clips quickly.

  • Enhance operating efficiency

Prior to packaging, agribusinesses must do a thorough inspection of their produce. To maximize shelf life, a tomato farmer, for example, could physically sort his tomatoes into six stages of ripeness, ranging from green to red as they mature. Because they don’t have to personally inspect each one, they can train a bespoke model that can detect tomatoes according to maturity. Automated tomato sorting and packing is possible by connecting the model to their production systems.


  • Simplify data labeling

Labeling photographs has never been easier thanks to the Rekognition Custom Labels console. An image can be labeled as a whole or specific objects can be selected using a simple click-and-drag interface.

  • Automated Machine Learning

Without a background in machine learning, it is possible to create custom models. Automatic machine learning, on the other hand, is enabled via Rekognition Custom Labels’ autoML capabilities. When photos are provided, it may then automatically import and evaluate the data, select the right machine learning algorithms for training, and provide model performance indicators.

  • Feedback and inference for simplified models

Determine how well your model performs. You can see the model’s prediction for each image in the test set compared to the label assigned to it. You can also look at f-scores, confidence scores, and precision/recall measures. It is possible to train your model with additional photos to improve its performance. Track your projections and correct any mistakes; then utilize feedback data to train new models and improve their results.

Dolan Cleary
Dolan Cleary

I am a recent graduate from the University of Wisconsin - Stout and am now working with AllCode as a web technician. Currently working within the marketing department.

Related Articles

Navigating AWS Complexity

Navigating AWS Complexity

Amazon’s Web Services is a very complex platform. Streamlining and optimizing production workflows can be challenging for inexperienced users. However, the benefit of learning grants options for better efficiency, reliability, security, and cost-effectiveness for operations run on AWS.

While complexity can be difficult to navigate, it’s not impossible. With the right level of expertise, AWS complexity can be navigated with ease.

What is Amazon Managed Grafana?

What is Amazon Managed Grafana?

Grafana stands out as a widely embraced open-source analytics and visualization platform, celebrated for its versatility in handling diverse data sources and delivering compelling dashboards and graphs. Renowned for its user-friendly interface, Grafana simplifies the process of data interpretation and enhances the overall experience by providing interactive visualizations.

AWS and re:Invent 2023

AWS and re:Invent 2023

There are plenty of AWS enthusiasts around the world such as ourselves with ideas on how to apply the Cloud in new and innovative ways. It’s a keynote where these enthusiasts come together, network, and share innovations and new methodologies with the public. Even for people less familiar with AWS, it is a great place to get first-hand experience with the platform either unguided or with professional help to see what opportunities the platform has.

Download our 10-Step Cloud Migration ChecklistYou'll get direct access to our full-length guide on Google Docs. From here, you will be able to make a copy, download the content, and share it with your team.