The effects of Pose on Facial Expression Recognition

Abstract

Research into facial expression recognition has predominantly been based upon near frontal view data. However, a recent 3D facial expression database (BU-3DFE database) has allowed empirical investigation of facial expression recognition across pose. In this paper, we investigate the effects of pose from frontal to profile view on facial expression recognition. Experiments are carried out on 100 subjects with 5 yaw angles over 6 prototypical expressions. Expressions have 4 levels of intensity from subtle to exaggerated. We evaluate features such as local binary patterns (LBPs) as well as various extensions of LBPs. In addition, a novel approach to facial expression recognition is proposed using local gabor binary patterns (LGBPs). Multi class support vector machines (SVMs) are used for classification. We investigate the effects of image resolution and pose on facial expression classification using a variety of different features.

Facial emotion recognition using empirical mode decomposition

“…Some of them have utilized pixel based (Rahulamathavan, Phan, Chambers, & Parish, 2013;Wang & Ruan, 2010), Gabor filter (Deng, Jin, Zhen, & Huang, 2005;Donato, Bartlett, Hager, Ekman, & Sejnowski, 1999;Owusu, Zhan, & Mao, 2014), wavelet transform (Kazmi, Qurat-ul-Ain, & Jaffar, 2012;Shih, Chuang, & Wang, 2008), facial contour (Gu, Venkatesh, & Xiang, 2010), edge and skin detection (Ilbeygi & Hosseini, 2012), discrete cosine transform (Gupta, Agrwal, Meena, & Nain, 2011;Kharat & Dudul, 2009) and local binary pattern (Feng, Hadid, & Pietik ainen, 2005;Liu, Yi, & Wang, 2009;Luo, Wu, & Zhang, 2013;Moore & Bowden, 2009;Shan, Gong, & McOwan, 2009;Zhang, Zhao, & Lei, 2012;Zhao & Zhang, 2011) have gained lots of successful experiences. Eventhough facial emotion recognitions have achieved a certain level of success, however the performance is far from human perception.…”

Facial Action Recognition Combining Heterogeneous Features via Multikernel Learning

“…Other features successfully used in AU detection are the Haar-like features with an adaboost classifier proposed by Whitehill and Omlin [33] or the Independent Components combined with SVMs by Chuang and Shih [34]. For the emotion recognition task, Moore et al [35] combine Local Gabor Binary Pattern (LGBP) histograms with SVM and Fasel et al [36] combine graylevel intensity with a neural network classifier. The challenge organizers provide a baseline method [37] where LBP are used to encode images, Principal Component Analysis to reduce features vector dimension and SVM classifier to provide either AU or emotion scores, depending on the task.…”

Mapping Forests: A Comprehensive Approach for Nonlinear Mapping Problems

“…Table 2 shows that MF is on a par with the state-of-theart and related works in both protocols of BU3DFE and Multi-PIE. In addition, [24] proposed an approach similar to PSC in [21] but that is based on a new descriptor called LGBP. They have reported 80.17% accuracy on Multi-PIE dataset with 7 viewpoints similar to Multi-PIE-P2 but using six expressions from 100 subjects.…”

“…To determine which forests to use for a new input sample, we rely on another regressor to predict the head pose of the face. Since the mappings are adapted to the pose of the input face, this approach yields significantly better results than using a single mapping [21,24]. For each specific pose, the input test sample is applied to the corresponding forests to explore the best mapping.…”