Building a Serverless Image Processing Pipeline with Amazon Rekognition

By Aditya Kumar

Imagine running an application where users upload thousands of images every minute. If you rely on traditional servers to process, analyze, and label those images, you quickly run into massive infrastructure bottlenecks and unpredictable costs.

The modern solution? Serverless architecture.

In this article, we will build an automated, event-driven image recognition pipeline on AWS. Every time a user uploads an image to an Amazon S3 bucket, a serverless workflow will trigger instantly — analyzing the image using AI, extracting labels, and storing the metadata securely.

What We Will Build

We’ll create a seamless pipeline using the following core AWS services:

• Amazon S3: For secure object storage and event triggering.
• AWS Lambda: To run our image-processing code entirely serverless.
• Amazon Rekognition: To automatically detect objects, scenes, and concepts within our images using machine learning.
• Amazon DynamoDB: A NoSQL database to store our processed image metadata with lightning-fast retrieval times.
• IAM & Bucket Policies: Enforcing the security principle of least privilege to keep our architecture locked down.

So how to start?

Create an IAM role, just type like this in AWS search bar:

Following window will appear once we open:

Click on the “roles” from the left-hand side menu: following window will appear

Select it like this: then click on the next button

Following window will appear:

Select the required policies, then click on next, following window will appear:

Write the name of the role, scroll down to see the permissions we had added, review it and then click on create role:

Following confirmation will occur:

Now, go to the S3 bucket:

Create a new bucket: following window will occur, fill it as shown below

Leave rest as default then click on the orange button at the bottom of the screen, following confirmation will occur:

Now, go to the search bar and search DynamoDB:

Following window will appear:

Click on the create table button, following screen will appear:

Fill it as shown below:

Keep rest as default and scroll down to click on the create table button, following confirmation will occur:

Wait till the table come in active status, like this:

Now, Open the lambda function:

Click on it following screen will appear:

Click on the “Create a function” button: following window will appear

Fill it as follows:

Change the Execution role:

Leave rest as default, then click on the orange button: following confirmation ill occur

Copy-paste the following code:

Then deploy the code, following confirmation will occur:

Now, let’s change the configuration of this lambda function: go to the configuration section

Click on the “edit” button: edit the timeout and save the changes

Now, we will edit the notification of the S3 bucket: go to the Properties section of the bucket

Scroll down and select “event notification”:

Click on the “create event notification”:

Also,

Scroll down and select the “destination” as shown:

Then click on the Save changes button: following confirmation will occur

Now, we will add an inline policy to the role we had created earlier:

Add the following line of Json:

Click next, and fill it as:

Following confirmation will occur:

Now, we will test it, but how?

Go to S3 bucket we had created earlier, and then upload an object over there:

Then click on the upload button:

Now, where can we find the output of the image uploaded, DynamoDB.

Contact Me 📧 Email: adii.utsav@gmail.com 🔗 LinkedIn: https://www.linkedin.com/in/aditya-kumar-3241b6286/ 💻 GitHub: https://github.com/Rememberful