Enhance content/asset search in AEM DAM/Azure Storage using FAISS vector database

[Hemant_arora]

Use Case:

Search for similar assets in AEM DAM/Azure Storage using the FAISS vector database. For example, searching for images classified as male.

This approach is particularly useful when metadata is unavailable for a given set of assets and duplicate/similar assets needs to be searched.

---

Overview

1. Fetch Images: Retrieve images from the repository.
2. Filter Images: Use face detection and gender classification to filter the images.
3. Extract Features: Convert the filtered images into feature vectors.
4. Index and Search: Index these feature vectors using FAISS and find similar images.

Prerequisites for Local Setup: An AEM instance with assets in DAM, Python, FAISS installed, and necessary Python libraries.

---

Models Used

• ResNet50: A deep learning model utilized for extracting features from images. The final classification layer is removed to use only the feature extraction part.

---

Steps

1. Fetch Images from AEM DAM
The script retrieves images from the AEM DAM.

2. Image Processing

• Face Detection: Use Haar cascade classifier, a machine learning object detection method, to detect faces.
• Feature Extraction: Convert images into feature vectors using the ResNet50 model.
• Gender Classification: For demonstration purposes, the gender is hard-coded as 'male'.

3. Indexing and Searching

• FAISS Index: Create an index of feature vectors using the FAISS library for similarity search.
• Querying: Query the index with the feature vector of the first image to find similar images.

4. Output Results

• The script displays the results of the similarity search, showing the nearest images and their distances from the query image.

---

Configuration in AEM

1. Add MIME Types: In the config manager, add content/*:image/png,image/svg+xml to org.apache.sling.security.impl.ContentDispositionFilter.
2. Update MIME Types: Modify the above expression and script to match the types of assets in DAM.

---

Execution

1. Run AEM Locally.
2. Upload Assets: Ensure assets with faces identified as male are uploaded.
3. Update the Script: Configure the DAM path, MIME type, and credentials in the script.
4. Set Environment Variable: Open CMD and set KMP_DUPLICATE_LIB_OK=TRUE.
5. Run the Script.

You can also create a Flask application to trigger this functionality via a button click or checkbox in AEM.

For details on deploying a Flask web app to Azure App Service, refer to the following guide:
Deploy a Flask Web App to Azure App Service

Output:

Hope you find it useful. Thanks for reading.

[MeenuSh]

This is really helpful, Thanks