All research
Paper Fingerprinting using Deep Learning
Robust material fingerprints for document-fraud detection.
A generative-embedding pipeline that produces robust fingerprints of paper textures to detect document fraud.
- Built an image-embedding pipeline for document-fraud detection using Beta-Variational Autoencoders.
- Used Shearlet transforms, VAEs and generative networks with a specialised perceptual loss for robust fingerprint generation.
- Generated a custom dataset of 55,000 paper-texture images for training.
- Benchmarked Gabor filters & Local Binary Patterns for robustness using Euclidean and Bhattacharyya distances.