Facial Expression Recognition for Low Resolution Images using Local and Global Features with SVM Classifier
Shubhangi Patil-Kashid, Y.M. Patil, Vijaya Rahul Pawar, Amruta Kashid · International Journal of Electrical and Electronics Engineering · 2023
Human Machine interaction similar to human-human interaction is possible if machines can recognize expressions on human faces during communication. Facial expression recognition is easy for humans but a difficult task for machines. Facial expression recognition (FER) by machines is expected in the next generation of computers. Image acquisition, preprocessing, feature extraction and classification are the steps in FER. The accuracy of recognition depends on all these steps. Happy, sad, angry, fearful, surprised, disgusted and neutral are the seven expressions considered for classification. Local Binary Pattern (LBP), Histogram of Orientations (HOG), Shift Invariant Feature Transform (SIFT) and Optical Flow (OPF) are the four different feature extraction methods that are used. This paper compares the different feature extraction methods, and multiclass SVM is used as a classifier. The experiment is performed on three different datasets. The novelty of the work is the low-resolution dataset and combination of local and global features. Local feature extractors LBP and HOG are more efficient than global feature extractors SIFT and OPF. Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) are used to reduce the dimensionality and improve the speed of execution. The confusion matrix is used to compute the accuracy of recognition, specificity and sensitivity. Recognition accuracy is suitable for images captured in controlled laboratory scenarios, but still, more work is required for wild images on both feature extraction and classifier.