CSL2050 - Pattern Recognition and Machine Learning

Course Project: Image Retrieval using Machine Learning

Team Members

• S Vara Prasad Reddy(B23CM1057)
• Yadav Karan Subhashchandra(B23EE1103)
• Kodakandla Manideep(B23CS1026)
• Priyanshu(B23CS1097)
• Neeraj Mansingh(B23CS1095)
• Kalkar Tejas Prasad(B23CM1051)

Project Description

This project is developed as part of the course CSL2050 - Pattern Recognition and Machine Learning, focused on solving the Image Retrieval problem using various machine learning techniques. The main objective is to classify images from the CIFAR-10 dataset and retrieve similar images based on the input query.

The dataset consists of 60,000 32x32 color images in 10 different classes, with 50,000 for training and 10,000 for testing. Since raw image data is not directly suitable for most ML algorithms, feature extraction was a major focus of our work. We explored multiple techniques including PCA, HOG, and QuickNet, but ResNet-based deep feature extraction produced the most informative and discriminative features.

Once the features were extracted, we trained multiple classifiers including:

• Logistic Regression
• Decision Tree
• Random Forest
• K-Nearest Neighbors (KNN)
• Naive Bayes
• Artificial Neural Network (ANN)
• SVM
• XGboost

Among all combinations, the ResNet + ANN pipeline achieved the highest accuracy and performance. This is because ResNet (a deep convolutional network) is capable of capturing high-level abstract features, which when passed to an Artificial Neural Network, helped it learn complex patterns and non-linear decision boundaries more effectively than traditional ML models.

Repository Structure

.
├── Cifar10Dataset/           # Original dataset files
├── Demo/                     # Contains code for demo app
├── Model/                    # Contains trained models and architecture definitions
├── PreProcessedData/         # Vectorized and transformed datasets (.pkl files)
├── prml_webpage/             # Contains code and assets for webpage 
├── EDA.ipynb                 # Exploratory Data Analysis on CIFAR-10
├── FeatureExtraction.ipynb   # Feature extraction using HOG, QuickNet, PCA, ResNet
├── Resnet.ipynb              # Deep feature extraction using ResNet
├── README.md                 # Project documentation
├── Report.pdf                # Final project report with complete analysis
├── presentation.pdf          # Project presentation slides

We used the CIFAR-10 dataset:

• 50,000 training images
• 10,000 testing images
• 10 categories including: airplane, automobile, bird, cat, deer, dog, frog, horse, ship, and truck.

---

Preprocessing & Feature Extraction

We explored and compared different feature extraction techniques:

• ResNet
• PCA (Principal Component Analysis)
• HOG + PCA
• QuickNet
• PCA + HOG
• HOG
• EDA

---

Models Evaluated

Each set of features was passed into multiple machine learning models:

• Artificial Neural Network (ANN)
• Bayesian Classifier
• Clustering
• Decision Tree
• Logistic Regression
• K-Nearest Neighbors (KNN)
• Random Forest
• SVM
• XGboost

---

Best Performing Configuration

ResNet + ANN achieved the best accuracy of 89%, and was used in the final deployed application.

---

Frontend Integration

A user-friendly frontend was developed to allow image uploads and display the top-5 most similar images.

Results confirm accurate retrieval based on learned features.

---

Demo

You can try the demo here

Running the App Locally

To run the app on your local machine, follow these steps:

# Navigate to the Demo directory
cd Demo

# Install the required dependencies
pip install -r requirements.txt

# Start the Streamlit app
streamlit run app.py

---

Objective

• Compare different feature extraction techniques
• Evaluate various machine learning models
• Integrate everything into an end-to-end image retrieval system

---