In this job, I collaborated with Chayma Bouzaidi

1. Overview
2. Requirements
3. Setting up MongoDB/GridFS
4. Start the App

In this project, we built a content-based image retrieval engine that allows users to send a request image in order to display similar images from 1M images database stored in GridFs (MongoDB).🧐

To build this app, we followed this main steps:
       • Building feature vectors using Principal Component Analysis
       • Storing feature vectors and thumbnail images in MongoDB
       • Building a VP-Tree to optimize searching time
       • Setting an interactive web app using Flask

NB : You can get 1M thumbnail images database and corresponding edge histogram descriptors from here

       • MongoDB (version 4.2.1).
       • Python (version 2.7.15+).
       • Create a MongoDB database called CBIR.
       • Specify your MongoDB connection URL in the script app/app.py.
       • Go to app/ directory and install the requirements via pip :
             • pip install flask numpy scipy matplotlib scikit-image gunicorn.
             • pip freeze > requirements.txt.

       • Go to scripts/ directory and run toGridFs.py script in order to dump the Thumbnails, the Principal Components and the VP-Tree into GridFs :
             • python toGridFs.py -td <YOUR_THUMBNAILS_DIRECTORY> -id <YOUR_PC_VPTREE_DIRECTORY>.
NB : You can get the Principal Components and the VP-Tree files from here and here

       • Run python app/app.py to start the App.
NB : The list of query images is static. You can change it by hosting the thumbnails on a Free server and specifying the image URL in img tag in app/templates/index.html.