Emotion Recognition

Project Definition:

Classifying a person’s emotion, from a list of 7 emotions, using their facial expression. Downloaded face images Kaggle Dataset that came as a csv file with grayscale images converted into binary, and used Haar Cascade Frontal Face code to detect faces in live video feed.

Emotions:

• anger
• disgust
• fear
• happy
• sad
• surprise
• neutral

Models & Metrics:

1. 4-layer model using relu and softmax --> 0.09 validation accuracy
2. 6-layer CNN model of tanh, MaxPooling2D, and softmax --> F1-score = 0.45
3. 6-layer CNN model of tanh, MaxPooling2D, and softmax, normalized data, batch-size = 10 early callback --> F1-score = 0.78
4. Added one more tanh layer to last model --> F1-score = 0.78
5. VGG16 with 2 tanh layers and softmax output layer --> F1-score = 0.88
6. VGG19 with 2 tanh layers and softmax output layer --> F1-score = 0.78