Table 1.
Notations and description.
Fig 1.
Binary neural network-based stacking ensemble model for pose-aware facial expression recognition.
Fig 2.
Samples of the 40 facial images associated with the five poses and eight facial expressions of one subject in the RaFD (Row wise, 00, 450, Frontal, -00, -450) facial expression from left to right (anger, happiness, surprise, sadness, fear, neutral, contempt, disgust).
Table 2.
Description of multi-pose facial expression database.
Table 3.
Optimal classifier structure for ensemble classifier.
Table 4.
Performance comparison of NB predictor with PCA and HOG features.
Table 5.
Front pose facial expressions recognition (%) accuracy with NB, KNN, SVM predictors, and PCA, HOG features.
Table 6.
Left (-45°) pose facial expressions recognition (%) accuracy with NB, KNN, SVM predictors, and PCA, HOG features.
Table 7.
Left (-90°) pose facial expressions recognition (%) accuracy with NB, KNN, SVM predictors, and PCA, HOG features.
Table 8.
Right (+45°) pose facial expressions recognition (%) accuracy with NB, KNN, SVM predictors, and PCA, HOG features.
Table 9.
Right (+90°) pose facial expressions recognition (%) accuracy with NB, KNN, SVM predictors, and PCA, HOG features.
Fig 3.
Principal component analysis (left) and histogram of oriented gradient (right) expression confusion matrix.
Fig 4.
The results of frontal pose confusion matrices comparing performance of NB, KNN and SVM on PCA and HOG features
: (a)-(c) confusion matrices correspond to PCA features, and (d)-(f) confusion matrices corresponds to HOG features.
Fig 5.
The results of pose -45° yaw pose confusion matrices comparing performance of NB, KNN and SVM on PCA and HOG features
: (a)-(c) confusion matrices correspond to PCA features, and (d)-(f) confusion matrices corresponds to HOG features.
Fig 6.
+ 45° yaw pose confusion matrices comparing performance of NB, KNN and SVM on PCA and HOG features: (a)-(c) confusion matrices correspond to PCA features, and (d)-(f) confusion matrices corresponds to HOG features.
Fig 7.
The results of pose -90° confusion matrices comparing performance of NB, KNN and SVM on PCA and HOG featur
es, (a)-(c) confusion matrices correspond to PCA features, and (d)-(f) confusion matrices corresponds to HOG features.
Fig 8.
+ 90° confusion matrices comparing performance of stacked ensemble model of type 1 combined with NB, KNN and SVM on PCA and HOG features: (a)-(c) confusion matrices correspond to PCA features, and (d)-(f) confusion matrices corresponds to HOG features.
Table 10.
Existing work on multi-pose facial expression recognition.
Table 11.
Classification accuracy of existing methods which used RadBoud Faces Database.