delongsh / rlecc

Face Template Protection Through Residual Learning Based Error-Correctinig Codes

This repository is code for our paper Face Template Protection Through Residual Learning Based Error-Correcting Codes.

Introduction

This code aims to implement a face template protection technique by using residual learning, which maps the facial images into low-density parity-check (LDPC) codewords. This implementation is based on PyTorch. In this paper, we use the LDPC coding algorithm developed by Radford M. Neal. The trained CNN model on extended Yale Face B with d = 256 can be downloaded.

Directory

Root

${ROOT}
|-- labels
|-- load_datasets
|-- models
|-- process_res
|-- split
|-- test_images
main-test.py
utils.py
requirements.txt

• labels: LDPC codewords for different face databases
• load_datasets: Load face images
• models: Define CNN architecture
• process_res: Process the predicted binary codes
• split: Store different orders for face database
• test_images: Store test images, download .

Database

The databases used in our paper contain extended Yale Face B, PIE and FEI Database.

Requirements

Install python packages

pip install -r requirements.txt

Running

python main-test.py -r 11-04-13-20 --seed 1 --dataset yaleB --ldpc-len 256 --dataset-seed 0 -t 38 0 --epochs 50 --batch-size 256

More Info

• Contact: Please send comments to [email protected]