Discriminative feature learning‐based pixel difference representation for facial expression recognition

Abstract

Recently, researchers have proposed different feature descriptors to achieve robust performance for facial expression recognition (FER). However, finding a discriminative feature descriptor remains one of the critical tasks. In this paper, we propose a discriminative feature learning scheme to improve the representation power of expressions. First, we obtain a discriminative feature matrix (DFM) based pixel difference representation. Subsequently, all DFMs corresponding to the training samples are used to construct a discriminative feature dictionary (DFD). Next, DFD is projected on a vertical two-dimensional linear discriminant analysis in direction (V-2DLDA) space to compute between and within-class scatter because V-2DLDA works well with the DFD in matrix representation and achieves good efficiency. Finally, nearest neighbor (NN) classifier is used to determine the labels of the query samples. DFD represents the local feature changes that are robust to the expression, illumination et al. Besides, we exploit V-2DLDA to find an optimal projection matrix since it not only protects the discriminative features but reduces the dimensions. The proposed method achieves satisfying recognition results, reaching accuracy rates as high as 91.87% on CK+ database, 82.24% on KDEF database, and 78.94% on CMU Multi-PIE database in the LOSO scenario, which perform better than other comparison methods.

Human-Computer Interaction Using Deep Fusion Model-Based Facial Expression Recognition System

“…Hence, the performance comparisons reported herewith have been made under the same training-testing protocol used by the proposed methodology. Table 6 shows the performance of analysis of Res-Net50 [57], Inception-v3 [58], Sun et al [59], and the proposed model on CK+ dataset. It is found that the proposed model achieves better performance with 96.89% performance.…”

“…It is found that the proposed model achieves better performance than the existing models by showing an average enhancement of 41.73% over the competitive models. [63] 74.05 10-fold CV with 720 images for 7 expression classes Inception-v3 [58] 75.04 10-fold CV with 980 images for 7 expression classes ResNet50 [57] 72.32 10-fold CV with 980 images for 7 expression classes Zavarez et al [64] 72.55 10-fold CV with 980 images for 7 expression classes Sun et al [59] 82.24 10-fold CV with 490 images for 7 expression classes Fard et al [61] 80.76 10-fold CV with 981 images for 7 expression classes Hussein et al [62] 81.87 10-fold CV with 981 images for 7 expression classes Proposed 83. 27 10-fold CV with 980 images for 7 expression classes 24.78 Liu et al [65] 26.14 Inception-v3 [58] 29.52 Dhall et al [66] 39.13 Fard et al [61] 33.45 Hussein et al [62] 37.31 Proposed 41.73…”

Mobile-Optimized Facial Expression Recognition Techniques

“…Generic features such as SURF have been used to describe facial features and perform FER with a good accuracy of up to 96% on the MUG dataset [21], [22]. 2D linear discriminant analysis can also obtain a good detection rate [20],…”

“…Texture features are significantly more computationally expensive than geometric features, but they provide a high accuracy. Some of the more popular approaches can be seen in [11]- [20]. Generic features [21], [22] use general-purpose descriptors.…”

Automatically Query Active Features Based on Pixel-Level for Facial Expression Recognition

“…To address this issue, two-dimension linear discriminant analysis (2DLDA) [27] making full use of structure information had been proposed to address the problem caused by above. Besides, Sun et al [33] presented a discriminative feature learning method based on vertical 2DLDA that fully considered the matrix format of data and time complexity. Besides, some subspace learning methods are also proposed to perform FER classification tasks [34]- [37].…”