Facial Expression Recognition in Videos Using Dynamic Local Binary Patterns
Akram Bafandehkar, Hamid Karimi-Rouzbahani · 2022
We propose a novel fully automatic approach for facial expression recognition in videos. For that purpose, we focus on face regions, such as eyes and mouth, which are mostly affected by variation in facial expressions. Initially, the face region is detected using Viola and Jones’ face detection method and the key parts of the face are located based on the key points extracted through pictorial structure analysis. Next, the image sequence is transferred to the motion vector domain using Lucas-Kanade algorithm. Later, the key parts of faces are separated using pictorial structures and a novel heuristic weighted Local Binary Pattern on Three Orthogonal Planes (LBP-TOP) is applied on each part to extract the most informative features. Finally, a combination of Support Vector Machine classifiers is used to classify facial expressions. The effectiveness of proposed method is evaluated on the Cohn-Kanade facial expression dataset. Our proposed method, , provided features that reached an average recognition accuracy of 98.48% in 10-fold cross-validation which improved the basic LBP-TOP method by 1.13%. Advantages of our approach is its robustness to rotation and gray-scale changes leading to high accuracy as well as being fully automatic.