Highlights

• Detection of objects, scenes, and activities

Amazon Rekognition Video recognizes hundreds of objects, scenes, and behaviors, such as delivering a delivery or dancing, in seconds. You get a confidence score for each label you identify. Bounding boxes are also provided for specific objects like ‘Person’ and ‘Car,’ allowing easier counting and localization. For complex operations like “blowing out a candle” or “extinguishing a fire,” Amazon Rekognition Video relies on motion in the video. It is possible to enhance the searchability of your content and deliver ads that are relevant to the content immediately preceding them by utilizing the detailed metadata provided by these rich media assets.

• Moderation of online content

For each detection, Amazon Rekognition Video offers a timestamp and a description of the flagged content. In addition, you’ll get a list of labels and confidence scores for each level of harmful content. “Graphic Female Nudity” is a sub-category of “Explicit Nudity,” for example. It is possible to put up different business rules for other markets and regions using confidence scores and detailed labeling.

• Detecting text

In addition to the text detection confidence, Amazon Rekognition Video offers the location bounding box and timestamp for each text detection. Filter words by ROI, bounding box size, and confidence score. For example, to read scoreboards in a soccer game, you may only want to detect text in the bottom third of the screen.

• Celebrity appreciation

It’s easy to spot famous faces in videos thanks to Amazon Rekognition Video. Celebrity names and unique ids are included, along with URLs to associated resources for the celebrity, including the celebrity’s IMDB link.

• Recognition and analysis of the face

Amazon Rekognition Video can locate the bounding box for up to 100 faces in a video frame. Gender, emotions, an estimate of the person’s age, and whether or not they are smiling may all be gleaned for each identified face, in addition to timestamps for each detection.

• Face-to-face search

When a video is played, Amazon Rekognition Video may identify familiar faces in it by searching against a private repository of face photos. A similarity score is assigned to each match, and timestamps are set to each instance where the same individual is detected throughout the movie. The Amazon Rekognition Video service can also group all unknown people in a video who do not have any matches in the repository and return timestamps with unique identifiers for each of these people.

• Pathing of individuals

If your video contains people, Amazon Rekognition Video can track where they are, when they are moving, and how they are moving. As a bonus, Amazon Rekognition generates a unique index for each individual it recognizes, allowing you to keep track of how many people are in the video.

• Video analysis of a live webcast

Using Amazon Rekognition Video, you can analyze your live video streams in real-time to detect and search for people. For example, using an Amazon Kinesis Video Streams stream as an input to Rekognition Video allows for a shallow latency face search against a library of your photographs by delivering a feed from Amazon Kinesis Video Streams.

Amazon Rekognition for Images

Deep learning plays a significant role in enhancing Amazon Rekognition’s capabilities by utilizing neural networks with multiple layers to mimic the human brain’s data processing methods. This approach enables Amazon Rekognition to delve deep into analyzing extensive datasets, learning from each piece of information to perform intricate tasks such as visual recognition, natural language processing, and predictive analytics. Specifically designed to handle unstructured data like images and videos, deep learning algorithms enable Amazon Rekognition to excel in identifying and categorizing visual content with high precision. By training on diverse datasets, these algorithms become adept at detecting objects, scenes, faces, understanding contextual information in images and videos, and even recognizing emotions conveyed through facial expressions. Thus, the integration of deep learning technologies equips Amazon Rekognition with the capability to provide advanced image and video analysis features, ensuring accurate results and facilitating a wide range of applications.

Highlights

• Identification of Objects and Scene

Rekognition Image recognizes thousands of things, such as automobiles, dogs, and furniture, and can provide accurate identification. Besides scenes within an image, such as a sunset or the beach, recognition may also detect objects within an image. As a result, you can quickly search through vast image collections while filtering and curating them.

• Recognition of Facial aspects

In an extensive collection of photos, Rekognition Image allows you to locate faces similar to your own. Then, using your photographs, you can generate an index of the faces that were discovered. Fast and accurate search results are returned by Rekognition Image, which produces faces that are the closest match to your reference face.

• Face Time analysis

Rekognition Image is a tool for identifying faces in images and determining whether or not facial expressions, such as a smile or open eyes, are present. In processing an image, Rekognition Image will return the position and a rectangle frame for every discovered face.

• Face Comparison

Rekognition Image allows you to determine how likely two faces in two images are the same person. Similarity scores can be used to check a user against a reference photo in real-time with Rekognition.

• Detection of insufficient content

Recognition images can detect explicit and suggestive information, allowing you to filter photographs according to your needs. In addition, Rekognition gives a hierarchical list of labels with confidence scores to enable fine-grained control over what photos you want to accept.

• Celebrity Appreciation

Rekognition Image recognizes and identifies thousands of renowned, notable, or influential people in their respective fields of work. This feature provides an index and search function for celebrity images in digital image libraries.

• Image with text

Text on road signs, license plates, t-shirts, and mugs, and text on screens like news headlines or captions, may all be found and extracted with Rekognition Image. There is a rectangle frame and a confidence score for each word or line that Text in Image detects in the image.

• The detection of Personal Protective Equipment (PPE)

Amazon Rekognition Image can determine whether or not a person in an image is wearing personal protection equipment (PPE), such as a face mask, hand gloves, and a cap (nose for face covers, head for head covers, and hands for hand covers).

• Management Through API, Console, or command line.

There are three ways to access Amazon Rekognition: the Amazon Rekognition API, the AWS Management Console, and the AWS command-line interface (CLI). Use the Rekognition APIs via the console, API, and CLI to find a face, detect labels and analyze faces. In addition, Amazon S3 and Amazon DynamoDB provide AWS Lambda blueprints for Rekognition that make it simple to start image analysis based on events in your AWS Cloud computing data stores.

• Legal Protection in the Office

Identity and Access Management (IAM) services are connected with Amazon Rekognition (IAM). Use IAM policies to regulate your account’s resource-level permissions and Amazon Rekognition API access.

• Human Consideration

You can quickly add a human review for dangerous image detection into Amazon Rekognition thanks to its direct integration with Amazon Augmented AI (Amazon A2I). Amazon A2I’s built-in picture moderation methodology allows Amazon Rekognition predictions to be readily evaluated and validated. An internal pool of reviewers can be used, or you can tap into a workforce of more than 500,000 independent contractors who are already doing machine learning jobs through Amazon Mechanical Turk with Amazon A2I. In addition, AWS computing pre-screens its workforce vendors to ensure that they adhere to quality and security standards. The Amazon A2I website and the Amazon A2I Integration with Amazon Rekognition in the Amazon A2I developer guide provide more information on creating human review workflows.

• Flexible Pricing Model

Amazon Rekognition provides a range of pricing options tailored to the specific needs of users based on the type and volume of analysis required. This flexibility extends to both image and video analysis services, where costs are determined by the quantity of data processed and the specific features utilized. By offering this adaptable pricing structure, Amazon Rekognition ensures accessibility to advanced visual analysis tools for businesses of varying sizes, eliminating the need for large upfront investments.

Amazon Rekognition Custom Labels

With Amazon Rekognition Custom Labels, you can recognize certain items and scenes in photos. For example, find your logo in social media posts, identify your items on shop shelves, sort machine parts in an assembly line, identify healthy and diseased plants, or find animated characters in films.

Creating a bespoke image analysis model takes time, knowledge, and resources and can take months. It also takes hundreds or tens of thousands of hand-labeled photos to offer enough data for the model to make reliable conclusions. It takes months to acquire this data and vast teams of labelers to prepare it for machine learning.

We do the heavy lifting with Amazon Rekognition Custom Labels. This add-on expands on Rekognition’s existing skills, trained on millions of photographs in various categories. Instead of thousands of photographs, submit a modest batch of training images (usually a few hundred or less) into our simple terminal. With pre-labeled photos, Rekognition can start training in seconds. If not, you may either name them yourself or use Amazon SageMaker Ground Truth to do it for you. Rekognition can prepare a custom image analysis model in just a few hours using your image data. A model is trained by loading and inspecting training data, selecting the appropriate machine learning algorithms, and providing model performance indicators. Your custom model can then be used in applications via the Rekognition Custom Labels API.

Use Cases

• Measuring brand coverage

Marketing agencies must accurately disclose their clients’ media coverage. They typically manually track client brands and items in social media, TV, and sports videos. With Amazon Rekognition Custom Labels, agencies can train a model to recognize client logos and items. Instead of manually tracking conventional and social media, they can run photos and video frames through the bespoke model.

• Find syndication content

Content creators must often sift through thousands of photographs and videos to locate relevant content for their shows. For example, sports broadcasters need to personally prepare highlight films about games, teams, and players for affiliates. Custom models can quickly generate a meaningful array of photos and clips by training them to recognize teams, players, and common game occurrences like goals scored, penalties, and injuries.

• Enhance operating efficiency

Agribusinesses must grade their produce before packing it. For example, a tomato grower may physically sort tomatoes into six maturity classes, from mature green to red, for the best shelf life. Instead of manually inspecting each one, they can train a bespoke model to identify tomatoes based on ripeness. They may sort and pack the tomatoes automatically by linking the model with their manufacturing systems.

Highlights

• Simplify data labeling

The Rekognition Custom Labels console provides an intuitive visual interface for labeling photos. Using a simple click-and-drag interface, you may label an entire image or select particular objects. 

• Automated Machine Learning

Custom models can be built without machine learning knowledge. However, autoML capabilities in Rekognition Custom Labels automate machine learning. It can then automatically import and evaluate the data, select the appropriate machine learning methods for training, and offer model performance metrics once the training images have been provided.

• Feedback and inference for simplified models

Perform a performance test on your unique model. On each image in the test set, you can see the model’s prediction vs. the label assigned. You can also examine precision/recall metrics, f-scores, and confidence scores. You can use your model or retrain it with more photos to increase performance. Follow up on your forecasts, rectify any errors, and use feedback data to retrain new model versions and enhance performance.

Related Articles