This program employs several face recognition algorithm along with 2D wavelet transformation and uses the model with best training performances. In almost all test runs, SVM was the most accurate method.
Functionality includes, annotating an image with recognized face, annotating entire video with recognized face, real time annotation of webcam stream.
Put the images of various persons on the dataset
folder. The folder name should be the name of the person.
Run create_dataset.m
to preprocess the image. Preprocessing crops the faces out and resizes the images.
Run train_model.m
to train models with SVM
, KNN
, Discriminant
algorithms. Training uses 2D wavelet transformation to extract features. The trained model is saved on the file models.mat
.
Use input_image.m
script to predict the faces on an image.
input_video.m
annotates all the recognized face on a video and writes a new annotated video.
live_detection.m
detects recognized face on live webcam stream.