This project titled “Emotion Recognition for Real-Time Feedback” performs facial expression analysis in near real-time from a live webcam feed. It classifies human expressions into 8 different classes (Happy, Sad, Angry, Contempt, Disgust, Fear, Surprise, Neutral).
Facial detection is carried out to obtain facial expression using an in-built python library followed by training of Support Vector Machine model. To solve this multi-class problem, Support Vector Machine uses several tuning parameters such as kernels, gamma and regularization. This model is trained on the dataset which comprises of 10,708 images.
Data mining algorithms such as Sampling and Ensembling techniques are used to improve accuracy of the trained model. SVM model obtained an accuracy of 78% against testing dataset. …..(dynamic accuracy)
Based on this project we have developed two working applications for obtaining real-time feedback.
- Song Detection:
The first application detects the genre of the song with the help of facial expression of the end-user while he/she is listening to song. - News Reading Expression:
The other application interprets the category of news by reading the facial expression of the user. Both these applications give us genuine real-time feedback.