I have recently completed a data science course at General Assembly in London and for the past 5 weeks in the evenings I have been working on my final project. Cognisant of the rising tide of fake news / fake images and more recently deep fakes on the internet I decided to create an image classifier that could spot whether pictures of faces had been photoshopped or not.
This project ended up bifurcating into two parts with part 1 concerning pixel-based semantic segmentation of images of faces and part 2 developing the classifier to detect fake faces. Part 1 is contained within this repo and part 2 can be found here.
A more detailed discussion of the methodology can be found in this blog post
The purpose of this mini-project was to familiarise myself with working with images and aim to create a pixel-wise semantic segmentation classifier for images of faces. This project utilises the dataset found in [1] and [2].
- Machine Learning (specifically random forest classifiers and K-Means clustering)
- Python
- jupyter
- Sklearn
- Scikit-image for image processing tasks
- Raw Data is kept here within this repo
- Data processing/transformation scripts are being kept here
- Notebooks are kept here
- If you want to contact me - reach out to me on LinkedIn
[1] Khalil Khan, Massimo Mauro, Riccardo Leonardi, "Multi-class semantic segmentation of faces", IEEE International Conference on Image Processing (ICIP), 2015 -- PDF
[2] Khalil Khan, Massimo Mauro, Pierangelo Migliorati, Riccardo Leonardi, "Head pose estimation through multiclass face segmentation", IEEE International Conference on Multimedia and Expo (ICME), 2017 In collaboration with YonderLabs -- PDF