Rekognition

🧠 What is Amazon Rekognition?

Amazon Rekognition is a fully managed computer vision service that enables you to extract information from images and videos using pre-trained deep learning models.

✅ It supports facial analysis, object detection, celebrity recognition, text detection (OCR), content moderation, and custom labels — all without needing to build and train your own ML model.

---

🧰 Key Use Cases

Use Case	Rekognition Feature
👁️ Face detection & analysis	Detect faces, emotions, age range, gender
🧍 Person tracking in video	Detect and track people frame by frame
🛡️ Content moderation	Flag unsafe or explicit content
🆔 Face search/recognition	Match faces against a collection
🔠 OCR (text in image/video)	Detect printed or handwritten text
📦 Object detection	Detect people, animals, vehicles, etc.
🎬 Video analysis	Use Rekognition Video for temporal data
🧠 Custom image classification	Train your own model using Rekognition Custom Labels

---

🔎 Feature Overview

Feature	Image	Video	Description
Detect Faces	✅	✅	Bounding box, age, emotions, gender
Recognize Celebrities	✅	✅	Prebuilt celeb face database
Detect Labels	✅	✅	Identify scenes, objects, activities
Detect Text (OCR)	✅	✅	Print or handwritten text from images/videos
Content Moderation	✅	✅	Flag adult/violent/inappropriate content
Custom Labels	✅	❌	Train on your own labeled dataset (no coding)
Face Comparison	✅	❌	Compare two faces or against a face collection
Person Pathing	❌	✅	Track individuals across frames in a video

---

🛠️ Rekognition vs SageMaker vs OpenCV

Task	Rekognition	SageMaker	OpenCV / Custom
Face & label detection	✅ Fully managed	❌ Custom only	❌ Needs own model
Custom classification	✅ No-code (Custom Labels)	✅ Full training pipeline	✅ From scratch
Video analytics	✅ Yes	❌ Not native	❌ Hard to scale
Serverless deployment	✅ Yes	❌ Needs infrastructure	❌ Manual setup

---

💵 Pricing Summary (2024)

Feature	Image Price (per image)	Video Price (per minute)
Label detection	$0.001	$0.12
Face detection	$0.001	$0.12
Face comparison	$0.0015	N/A
Custom Labels training	$1.00/hour (training)	N/A
Custom Labels inference	$4.00 per 1,000 images	N/A
Content moderation	$0.001	$0.10

🧠 Tip: For large datasets, use batch processing (e.g., with S3 and Lambda) to reduce cost.

---

🧪 Sample Python Code (Boto3)

1. Detect Labels in an Image

import boto3

rekognition = boto3.client('rekognition')

response = rekognition.detect_labels(
    Image={'S3Object': {'Bucket': 'my-bucket', 'Name': 'my-image.jpg'}},
    MaxLabels=5,
    MinConfidence=80
)

for label in response['Labels']:
    print(f"{label['Name']} - {label['Confidence']:.2f}%")

---

2. Face Comparison

response = rekognition.compare_faces(
    SourceImage={'S3Object': {'Bucket': 'my-bucket', 'Name': 'face1.jpg'}},
    TargetImage={'S3Object': {'Bucket': 'my-bucket', 'Name': 'face2.jpg'}},
    SimilarityThreshold=80
)

for faceMatch in response['FaceMatches']:
    print("Similarity:", faceMatch['Similarity'])

---

3. Detect Text in Image

response = rekognition.detect_text(
    Image={'S3Object': {'Bucket': 'my-bucket', 'Name': 'receipt.jpg'}}
)

for text in response['TextDetections']:
    print(text['DetectedText'], text['Confidence'])

---

📦 Rekognition Custom Labels (No-code ML)

Train your own model using S3-labeled images:

• Step 1: Label images via Ground Truth or CSV manifest

• Step 2: Create a Rekognition Custom Labels Project

• Step 3: Train the model (serverless training)

• Step 4: Deploy the model endpoint and invoke via API

⚠️ Note: Custom Labels pricing is different (more expensive, but very powerful)

---

🔐 Security and Access Control

Feature	Details
IAM Role	Required to use Rekognition APIs
S3 Integration	Uses S3Object for media input
KMS Encryption	Supported if S3 bucket uses SSE-KMS
VPC Support	Not applicable (Rekognition is regional API only)

---

🛠️ Terraform Integration (Basic)

Amazon Rekognition does not have extensive Terraform support (e.g., no aws_rekognition_custom_label), but you can:

• Manage IAM roles/policies

• Automate image pipelines (S3 + Lambda + Rekognition)

• Use AWS Lambda to invoke Rekognition API programmatically

Example: S3 + Rekognition + Lambda can be orchestrated via Terraform for event-driven analysis.

---

✅ TL;DR Summary

Feature	Amazon Rekognition
Pre-trained Models	✅ Yes
Custom Models	✅ Rekognition Custom Labels
Real-time Video	✅ Yes (via Kinesis Video Stream)
Face Search/Compare	✅ Yes
OCR	✅ Yes (text detection in images/videos)
Content Moderation	✅ Yes
Serverless	✅ Fully managed APIs
Terraform Support	⚠️ Limited
Ideal For	Fast, no-ML-code visual recognition tasks

---

🔄 Popular Integrations

Service	How It Works
S3	Store and analyze media (image/video)
Lambda	Trigger analysis on file upload
Step Functions	Build image moderation pipelines
SNS/SQS	Notify users based on Rekognition results
SageMaker	Extend with more complex model use cases

---