bavaria95/thesis

Thesis Project

Face emotion recognition using convolutional neural networks

Aim of this project was to create a simple library to recognize one of 7 emotions (anger, contempt, disgust, fear, happy, sadness, surprise) of the person on the image using either pre-trained network by us or train it by yourself firstly.

Dataset

As base algorithm for this project Convolutional Neural Networks have been used (modification of LeNet architecture, with dropout technique).

Using

In order to use pre-trained network you can load saved network

network = nt.build_cnn('models/model.npz')

load image

faces = net_training.load_img("images/anger.jpg")

and evaluate this image with a given network

tab = net_training.evaluate(network, faces)

Optionally, if you want to train network by yourself you need to create a certain folder structure:

• Data
  • Images
    • 1 -- photo1.jpg -- photo2.jpg
    • 2 -- photo1.jpg -- ...
    • ...
  • Labels
    • 1 -- label.txt
    • 2 -- label.txt
    • ...

and execute function to train by providing specified paths

net_training.train_net(datadir, imagedir, labeldir, network)

Examples

anger = 100.00%
contempt = 0.00%
disgust = 0.00%
fear = 0.00%
happy = 0.00%
sadness = 0.00%
surprise = 0.00%

anger = 0.49%
contempt = 0.02%
disgust = 97.62%
fear = 1.47%
happy = 0.00%
sadness = 0.40%
surprise = 0.00%

anger = 0.00%
contempt = 0.00%
disgust = 0.00%
fear = 0.00%
happy = 0.00%
sadness = 0.08%
surprise = 99.92%

anger = 0.00%
contempt = 0.00%
disgust = 14.98%
fear = 0.00%
happy = 0.00%
sadness = 85.02%
surprise = 0.00%