抽象的

Recognition of Facial Expressions in Image Sequence using Multi-Class SVM

Geethu G S, Kamatchi T

Facial expression conveys the emotional state of an individual to observers, which is in the form of nonverbal communication. Recognition of facial expression plays a vital role in the field of Human Machine Interfaces (HMIs). Most of the existing automated system regarding facial expression has an impact over recognition rate. The seven facial expressions used in this work are happy, surprise, sad, fear, anger, disgust, and neutral. This paper proposes the Multi-class SVM to obtain high accuracy for recognizing each facial expression. The face in an image sequence is detected using Viola-Jones algorithm. Geometric and Appearance based feature extraction is used here. The geometric features are extracted using Image normalization and Thresholding techniques and optimal feature points are selected by calculating the entropy. Local Directional Pattern (LDP) + Local Directional Pattern variance (LDPv) descriptor helps to extract the appearance based features which contains the edges, spots and corners of a facial image. Finally, the performance of the proposed classifier is evaluated. Cohn-Kanade database is used to train and test the facial expression recognition system.