UCLA

Machine learning and deep learning techniques currently are leveraged by various practi- tioners in computer vision field. This thesis proposes an application of two different machine learning models to classify facial expressions of human and replace them with emojis accord- ingly.

The first model is based on fisher-face algorithm which is used for face recognition, the second one is a simplified version of prevalent Convolution Neural Network model utilized by many researchers: VGGNet network. The former model is implemented on OpenCV library, whereas convolutional neural network is builded with help of Keras. I trained both models by an open source dataset (JAFFE) which divides its facial expression images into seven classes. Both models are tested with randomly selected images of facial expression from internet.