Bag of Visual Words Image Classifier
This Git repository contains three image classifiers that utilizes the Bag of Visual Words (BoVW) approach with three different classification algorithms implemented from scratch: Naive Bayes, Logistic Regression, and a Fully Connected Neural Network.
The Bag of Visual Words (BoVW) is a popular technique in computer vision for image classification tasks. It involves breaking down an image into smaller regions, extracting local features from these regions by SIFT (Scale Invariant Feature Transform), and then creating a histogram of visual words (clusters of features) to represent the image. In this repository, we implement a BoVW-based image classifier with three different classification algorithms:
- Naive Bayes: A probabilistic classification algorithm based on Bayes' theorem, which assumes independence between features.
- Logistic Regression: A linear classification algorithm that models the probability of a sample belonging to a particular class.
- Fully Connected Neural Network: A multi-layer perceptron (MLP) neural network for non-linear classification.
These algorithms are implemented from scratch to provide a deeper understanding of their inner workings. The histograms can be found under data folder.