Recognize is an application which can be used for facial recognition. It includes a server which can listen for images and return a name prediction. This was done as the final project for EECS 445 for the Fall 2015 semester at the University of Michigan.

This application requires you to install the following software packages:

The code contained in this repository can be used for both training a neural network to learn features of faces and as a server which predicts names based on the previously trained model. If you do not wish to train your own network, you can download and use the pre-trained VGG-Face model. At this point, you simply need to run recognizeServer.m in order to start the server.

alignCrop - this code is used for frontalization, alignment, and normalization of the face for our neural network (not VGG-Face).

extractFeatures - given the VGG-Face neural network and the image, this function will extract the 1x4096 feature vector.

loadFacialFeatures - applies detectCrop to each roster image (roster at face_dir) and uses the neural net given in order to extract features from each.

loadNetwork - loading function that loads up and properly modifies the VGG-Face neural network (assumes it is at toolbox/models/vgg-face.mat) (doesn't load it if already in the workspace), generates names and feature vectors from the roster folder (or just loads them in if roster_features.mat, roster_names.mat exists in current working directory).

neuralNetwork - matconvnet description of our original (large) neural network.

neuralNetworkSmall - matconvnet description of the network we decided to use for training.

normalize - function to normalize an image.

predictName - given a normalized image of a face, this function is used to predict its owner's name.

recognizeServer - loads up a network and starts to listen for requests from an application on port 9090 and the public server IP. Replies with the prediction when done.

roster_features, roster_names - precomputed features and names for the EECS445 roster

testNetwork, trainNetwork - functions to test and train our neural network

A Google Glass application which makes use of this server can be found here.

#Contributors Kyle Balousek, Oleks Tyberkevych, Joyce Zhang, Haizhou Zhao