ujjwalaananth / CrossAgeFaceRecognition

Exploring a way to verify/recognize faces with variation in age

CrossAgeFaceRecognition

Exploring a way to verify/recognize faces with variation in age.

How to run:

Download combined_features.csv, combined_targets.csv and runclass.py. Navigate to file location, and give command

$python runclass.py

issues to resolve:

1. Feature Extraction Code

Currently, features are extracted from image dataset using MATLAB code of CARC (Cross-Age Reference Coding). At the moment, features must be extracted and saved separately before applying classifiers. In progress: converting MATLAB code to python OR findng a way to run MATLAB code within python code and save features dynamically.

2. Basic Dataset

The dataset that we are currently using shows little to no variation in pose or illumination. Classifier results will be closer to real-world results on across-age dataset with more variation

next steps:

1. New dataset

Currently we are gathering more varied cross-age dataset to test our classifier on

2. Finer feature extraction

Using architecture similar to that of Facebook's DeepFace network, we can improve accuracy by extracting finer features from given image. These features, in conjunction with CARC features, might result in a hihgly accurate cross-age recognition/verification system.