Real-Time Classification of Facial Expressions Using a Principal Component Analysis and Convolutional Neural Network

Abstract

Classification of facial expressions has become an essential part of computer systems and human-computer fast interaction. It is employed in various applications such as digital entertainment, customer service, driver monitoring, and emotional robots. Moreover, it has been studied through several aspects related to the face itself when facial expressions change based on the point of view or perspective. Facial curves such as eyebrows, nose, lips, and mouth will automatically change. Most of the proposed methods have limited frontal Face Expressions Recognition (FER), and their performance decrease when handling non-frontal and multi-view FER cases. This study combined both methods in the classification of facial expressions, namely the Principal Component Analysis (PCA) and Convolutional Neural Network (CNN) methods. The results of this study proved to be more accurate than that of previous studies. The combination of PCA and CNN methods in the Static Facial Expressions in The Wild (SFEW) 2.0 dataset obtained an accuracy amounting to 70.4%; the CNN method alone only obtained an accuracy amounting to 60.9%.

Multilabel image analysis on Polyethylene Terephthalate bottle images using PETNet Convolution Architecture

“…CNNs have a layered architecture with shared weights, sparse connections, and pooling operations that can identify both short-term and long-term patterns occurring in different parts of the time series [17]. However, CNN has produced numerous new breakthroughs in various applications, including segmentation, object recognition, and detection [18].…”

Generating images for Supervised Hyperspectral Image Classification with Generative Adversarial Nets

“…Many supervised learning models have been explored in the space of hyperspectral image classification. Before the recent success of Deep Neural Network (DNN) seen in areas of computer vision such as image classification [11], traditional supervised learning algorithms such as K Nearest Neighbor (KNN) and maximum likelihood were used [1].…”

“…On the other hand, there is a limited number of training samples in HSI classification due to the cost and intensive labor required to gather and label data [12]. Although there has been progress in using the DL model for HSI classification, the lack of training data makes DL models prone to overfitting, that is, performing well on the training dataset but showing suboptimal performance in the test dataset [11]. Hence, it is imperative to devise ways to largen existing datasets as it is critical to the performance of DL models.…”

Human Facial Expressions Identification using Convolutional Neural Network with VGG16 Architecture

“…Humans can produce different facial expressions [1], but some distinctive facial configurations are associated with specific emotions [2], regardless of gender [3], age [4], cultural background [5], and socialization history [6]. Facial expressions accounted for 55% of message delivery, while language and voice accounted for 7% and 38%, respectively [7].…”