Proposal for final project

For the final project, I want to create an intelligent selfie filter based on user emotion. Normally, a user has to select a filter style before or after taking a selfie. What if a user is able to take a selfie, and the intelligent selfie filter is able to recognize their emotions (facial expressions or gestures) and generates a corresponding filter style based on the user emotion.

The techniques:
Emotion Detection: a deep neural network will be used to classify some pre-defined emotion labels, such as happy, sad, angry, excited, etc. A large amount of emotion training data needs to be fed into the classifier. To ease the classification, some potential gesture landmarks could be used to disambiguate between subtle facial expression. A reasonable accuracy needs to be achieved for the following filter application.

Style Transfer: pre-trained style model will be saved and selected based o the emotion classification result. If a happy face is detected, the happy style will be applied to the live camera video feed and the user could observe the intelligent filter response. The user will be able to capture and save the filter result.