In today’s fast-paced world, businesses need efficient and innovative solutions to stay ahead. For one of our clients, the challenge was clear: searching through a massive inventory of over 2,000 yarn designs to match a customer’s request. Traditional methods were time-consuming, often leading to delays or even missed opportunities. That’s when we stepped in to create a state-of-the-art image recognition app, powered by AI and AWS, to streamline their operations and deliver results faster.

Understanding the Use Case

Our client’s primary requirement was to quickly and accurately identify similar yarn designs from their inventory based on an image provided by customers. The existing manual process was inefficient, prone to errors, and negatively impacted customer satisfaction.

The solution? An AI-powered mobile app that could:

• Accept a customer’s yarn design image.
• Instantly analyze and match it with similar designs in the inventory.
• Deliver accurate results, saving time and boosting productivity.

The Technical Stack

To bring this vision to life, we leveraged a combination of advanced AWS services and machine learning technologies:

• AWS Rekognition: For initial image analysis and feature extraction.
• AWS Lambda: To handle serverless computing needs.
• Amazon S3: For storing images and results.
• Amazon SageMaker: To build and train custom machine learning models for better accuracy.
• Amazon Elasticsearch Service: To enable efficient search and retrieval of results.
• Amazon SQS: To manage asynchronous and long-running tasks.

Building the Image Recognition System

1. Initial Integration with AWS Rekognition

We began the journey by utilizing AWS Rekognition for its pre-built image analysis capabilities. Here’s how the system was set up:

• Lambda Function: A Lambda function was created to handle image uploads and invoke the Rekognition API. This function:
  • Processed uploaded images.
  • Analyzed them using Rekognition.
  • Stored the results in Amazon S3 for easy access.
• API Gateway: To provide a seamless interface for the app, an API Gateway was configured to trigger the Lambda function.

2. Overcoming Challenges

While AWS Rekognition provided a good starting point, several challenges arose:

• InvalidImageFormatException: Some uploaded images were in unsupported formats. We resolved this by validating and converting images to formats like JPG and PNG.
• Request Timeouts: Processing large images within API Gateway’s default 30-second timeout was difficult. To address this, we integrated Amazon SQS to manage long-running tasks asynchronously.

3. Introducing SQS for Scalability

With Amazon SQS, the workflow was redesigned to handle tasks more efficiently:

• Image processing requests were sent to an SQS queue.
• A secondary Lambda function was created to process messages from the SQS queue.

This setup ensured that long-running tasks did not disrupt the user experience, as the system could now handle multiple requests simultaneously.

4. Transition to Custom Models with SageMaker

While AWS Rekognition offered out-of-the-box features, we needed more customization for accurate image matching in our client’s specific domain. Using Amazon SageMaker, we:

• Built and trained custom machine learning models tailored to the client’s inventory.
• Integrated these models with Elasticsearch for efficient search and retrieval of matching results.

5. API Development

To provide a user-friendly interface for the app, we developed two key APIs:

• POST /analyzeImage: To initiate the image analysis and return a request ID for tracking.
• GET /results/{request_id}: To retrieve analysis results based on the request ID.

6. Handling Dependencies in Lambda

To include external libraries like scikit-learn for advanced processing, we created a Lambda Layer:

• Packaged the required libraries into a zip file.
• Uploaded it to Lambda Layers.
• Attached the layer to our Lambda function for seamless execution.

Results Achieved

The AI-powered app revolutionized our client’s operations by:

• Reducing search time from hours to seconds.
• Improving accuracy in identifying similar yarn designs.
• Enhancing customer satisfaction with quick and reliable results.

Key Takeaways

This project highlights the power of combining AI and cloud technologies to solve real-world problems. With AWS services like Rekognition, SageMaker, and SQS, we were able to build a scalable, efficient, and highly accurate image recognition system.

If your business is looking to harness the power of AI and cloud computing to streamline operations, improve customer satisfaction, and drive innovation, we can help. Contact us today to learn more about how we can transform your ideas into impactful solutions.