Local Directional Pattern Variance (LDPv): A Robust Feature Descriptor for Facial Expression Recognition

Mohammad Kabir, Taskeed Jabid, and Oksam Chae
Department of Computer Engineering, Kyung Hee University, South Korea

 

Abstract: Automatic facial expression recognition is a challenging problem in computer vision, and has gained significant importance in the applications of human-computer interactions. The vital component of any successful expression recognition system is an effective facial representation from face images. In this paper, we have derived an appearance-based feature descriptor, the Local Directional Pattern Variance (LDPv), which characterizes both the texture and contrast information of facial components. The LDPv descriptor is a collection of LDP codes weighted by their corresponding variances. The feature dimension is then reduced by extracting the most discriminative elements of the representation with Principal Component Analysis (PCA). The recognition performance based on our LDPv descriptor has been evaluated using Cohn-Kanade expression database with a Support Vector Machine (SVM) classifier. The discriminative strength of LDPv representation is also assessed over a useful range of low resolution images. Experimental results with prototypic expressions show that the LDPv descriptor has achieved a higher recognition rate, as compared to other existing appearance-based feature descriptors.

Keywords: Facial expression recognition, feature descriptor, Local Directional Pattern (LDP), LDP variance (LDPv), Principal Component Analysis (PCA), and Support Vector Machine (SVM) classifier.

  

Full Text