Facial Expression Recognition using Hand-Crafted Features and Supervised Feature Encoding

Facial Expression Recognition (FER) is an emerging field of research because of the latest advancement in the computing field making the user's experience a clear priority. The proposed framework focuses on facial expression recognition using state-of-the-art hand-crafted feature extraction algorithms. First, pre-processing of images is applied to reduce noise and faces are then extracted using Viola Jones algorithm. Then, salient features are extracted from the region of interest using different feature extraction methods. We compare the performance of methods such as Speed-Up Robust Features (SURF), Features from Accelerated Segment Test (FAST), Histogram of Oriented Gradients (HOG) and Harris Corner Detection. The high-dimensional extracted features are quantized using supervised k-mean clustering. Finally, the proposed method is evaluated on the publicly available CK+ dataset and the model suggested achieves 90.79 per cent predictive accuracy. Angry, disgust, fear, happy, sadness, surprise, contempt are the seven different emotions to be recognized in CK+ dataset.