My computer vision coursework submitted at the University of East Anglia.
The goal of this project is to correctly categorise images using image feature extraction techniques and machine learning algorithms.
The project classifies scenes from a subset of the SUN database in to 1 of 15 categories: Kitchen, Store, Bedroom, LivingRoom, House, Industrial, Stadium, Underwater, TallBuilding, Street, Highway, Field, Coast, Mountain, Forest.
Image features extraction techniques used:
- Tiny Images
- Colour Histogram
- Bag of SIFT + colour
- Bag of HOG2X2
- Spatial Pyramid Kernel SIFT + colour
- Spatial Pyramid Kernel HOG2X2
Machine learning algorithms used to classify the image features:
- K Nearest Neighbour
- Support Vector Machine
An example result using the Spatial Pyramid Kernel HOG2X2 with the Support Vector Machine.
Accuracy: 76.5%
Confusion matrix:
This project requires VLFeat to run.
Download binary package at: http://www.vlfeat.org/download.html
Then link VLFeat to the project in coursework2_starter.m at line 74:
74: run('{your_file_path}/vlfeat-{version}/toolbox/vl_setup');