From the course: Developing Modern Applications with AWS AI and Generative AI Services
Amazon Rekognition - Amazon Web Services (AWS) Tutorial
From the course: Developing Modern Applications with AWS AI and Generative AI Services
Amazon Rekognition
- [Instructor] Let's explore how to process images and videos using AWS AI services, beginning with an in-depth exploration of Amazon Rekognition. Amazon Rekognition is an AI-powered image and video analysis service that helps automate computer vision tasks like object detection, facial recognition, content moderation, and text extraction from images and videos. It enables developers to achieve this without needing deep expertise in machine learning. Let's see what are its core features? The first is object and scene detection, which detects scenes, objects, and celebrities in images and videos. Text detection detects and recognizes handwritten text in images and videos in many different languages. Next is content moderation, which detects and filters inappropriate and explicit content. Next is personal protective equipment detection, which allows businesses to detect the presence of specific equipment or gear that the employees must wear during their work. Next is facial analysis, which detects, analyzes, and compares facial attributes like gender, age, and emotions in images, streaming, and stored videos. Finally, people pathing that tracks identified people as they move across video formats. Now that we have seen the core features, let's see how Amazon Rekognition works. It starts with uploading the images or videos to be processed, preferably to an Amazon S3 bucket. These images or videos are fed into pre-trained deep learning models to detect objects, faces, text, and unsafe content. The service returns labels or alerts based on the analysis. Amazon Rekognition offers two special features. The first one is custom labels. It starts again with uploading the images and videos to the S3 buckets. Then users label the images with custom labels they want to detect. It's better to have both positive and negative examples. Amazon Rekognition processes and trains a custom model based on the provider images. The trained model is tested for accuracy, prediction, and recall. Once trained, the model can no longer analyze new images to detect the custom objects. This feature allows businesses to train machine learning models for image and video analysis tailored to their specific needs, such as detecting sponsor logos in sports footage, brand specific logos in fashion designs, or any patterns that are not included in pre-trained models. The next feature is custom moderation, which allows businesses to create custom content moderation models, like custom labeling. It starts with uploading images or videos to the S3 bucket. The users must label their images by identifying the content they want to be detected. It could be a counterfeit product image, misleading ad, offensive symbols, etc. Amazon Rekognition automatically trains a model based on the provider dataset. The model's accuracy, precision, and recall are assessed. Once trained, the model can analyze new images and videos, flagging content that violates custom moderation rules. This feature lets businesses detect content that are specific to their policies, brand, or industry requirements. Some of the common use cases include detecting defective products in manufacturing and quality control, automatically cataloging the product in retail and e-commerce, detecting brand placements in media and entertainment, and finally detecting unauthorized personnel with no safety gear in the field of production and manufacturing.
Download courses and learn on the go
Watch courses on your mobile device without an internet connection. Download courses using your iOS or Android LinkedIn Learning app.