A Review of Facial Emotion Recognition: Comparison of Conventional and Deep Learning Methods
This paper deals with a comparison study including an update and an overview of techniques used for facial emotion expressions. First, we discuss the various models used for facial emotion representation also we describe the process of feature extraction. Along with that, we present Conventional approaches used for facial emotion recognition as well as deep learning approaches. In the paper, we define the advantages of Conventional and Deep Learning approaches and the drawbacks of both approaches. The paper points out also the challenges of both approaches which can be a point of interest for a Ph.D. study.
Deep Convolutional Neural Network for Robust Facial Emotion Recognition
Emotion and the ability to understand them are considered a channel of non-verbal communication. It is an important factor to achieve a smooth and yet robust interaction between machines and humans. In this paper, we review CNN-based methods for facial emotion recognition and we propose a new cutting edge deep learning approach to classify facial expressions from pictures. To guarantee the efficacy of the method, we used multiple datasets: FER2013, AffectNet, RaFD, and KDEF. We obtained results respectively 82.3%, 76.79%, 78.58 %, and 77.08 %. Those results surpassed the current state of the art. We also compared our achieved measurements to available APIs for facial emotion recognition.
A Review of Activation Function for Artificial Neural Network
The activation function plays an important role in the training and the performance of an Artificial Neural Network. They provide the necessary non-linear properties to any Artificial Neural Network. In this work, we provide a review of the most common and recent activation functions used in the hidden layer of an Artificial Neural Network.
Application and Perspectives of Convolutional Neural Networks in Digital Intelligence
Convolutional neural networks are the state-of-the-art approach for advanced computer vision tasks, as they offer capabilities beyond the straightforward application for image processing. This review provides an introduction to five areas where convolutional neural networks are a core topic of research: 2D and 3D object classification, image segmentation, few-shot learning, reinforcement learning, and explainability of neural networks. Each section provides an introduction to the research topic, identifies the main research questions, and lists modern solutions to these problems.
What makes a smile? A Deep Neural Network Point of View
Artificial intelligence is the mainstream solution to various problems, and thanks to developments in hardware, it is possible to achieve performance like never before. Continuous data collection provides us with opportunities for the creation of various datasets, which are the basis for various challenges. One of those challenges is recognizing emotions from humans’ facial expressions. Multiple deep learning models exist in the wild to solve such a task. They always yield high accuracy on their respective validation and test set. However, the performance of such a model tends to decrease when used on real-world images. This work gives insight into how such a deep learning model can predict facial expressions from biases present in training data.
Explaining Deep Neural Network using Layer-wise Relevance Propagation and Integrated Gradients
The field of artificial intelligence is the subject of research by a wide scientific community. In particular, through improved methodology, the availability of big data, and increased computing power, today’s machine learning algorithms can achieve excellent performance that sometimes even exceeds the human level. However, due to their nested nonlinear structure, these models are generally considered to be “Black boxes” that do not provide any information about what exactly leads them to provide a specific output. This raised the need to interpret these algorithms and understand how they work as they are applied even in areas where they can cause critical damage.
3D Object Classification Using HOG3D
A crucial component of artificial intelligence and image processing is 3D object classification. It helps to achieve significant and complex changes in performance through feature representation and processing of the images. Feature extraction plays a significant step in machine learning as it facilitates the feeding of insightful and non-redundant values to the machine learning algorithms. In this paper, we will present a framework to construct a 3D object classification system using several machine learning classifiers, and features were extracted using a local object structure descriptor called the 3D Voxel histogram of oriented gradient. We say that incorporating 3D classification tasks is a powerful strategy. This means enhancing the performance, precision, and efficiency of learning. The system contributed to increase efficiency and produced impressive results of 88 and 89% accuracy using Support vector machine and extreme Gradient Boosting, respectively. The results will be discussed and evaluated.
Integration of 2D Textural and 3D Geometric Features for Robust Facial Expression Recognition
Recognition of facial expressions is critical for successful social interactions and relationships. Facial expressions transmit emotional information, which is critical for human-machine interaction; therefore, significant research in computer vision has been conducted, with promising findings in using facial expression detection in both academia and industry. 3D pictures acquired enormous popularity owing to their ability to overcome some of the constraints inherent in 2D imagery, such as lighting and variation. We present a method for recognizing facial expressions in this article by combining features extracted from 2D textured pictures and 3D geometric data using the Local Binary Pattern (LBP) and the 3D Voxel Histogram of Oriented Gradients (3DVHOG), respectively. We performed various pre-processing operations using the MDPA-FACE3D and Bosphorus datasets, then we carried out classification process to classify images into seven universal emotions, namely anger, disgust, fear, happiness, sadness, neutral, and surprise. Using Support Vector Machine classifier, we achieved the accuracy of 88.5% and 92.9% on the MDPA-FACE3D and the Bosphorus datasets, respectively.
Experimental Evaluation of Cloud-Based Facial Emotion Recognition Services
The main goal of this paper is to perform an extensive analysis of the accuracy of six selected cloud-based facial emotion recognition services on three facial images datasets. The evaluation was performed on more than 33 000 images depicting eight different emotions. Results show that emotion recognition services show a varying level of accuracy over different types of datasets, having a lower accuracy for images of lower quality, but performing considerably better for images taken in ideal conditions. Based on these results we believe that cloud-based facial emotional recognition services do not have the expected accuracy for some use cases and therefore must be selected with care when developing a system that relies on emotion-based interactions.
A Large-Scale Study of Activation Functions in Modern Deep Neural Network Architectures for Efficient Convergence
Activation functions play an important role in the convergence of learning algorithms based on neural networks. They provide neural networks with the nonlinear ability and the possibility to fit in any complex data. However, no deep study exists in the literature on the comportment of activation functions in modern architecture. Therefore, in this research, we compare the 18 most used activation functions on multiple datasets (CIFAR-10, CIFAR-100, CALTECH-256) using 4 different models (EfficientNet, ResNet, a variation of ResNet using the bag of tricks, and MobileNet V3). Furthermore, we explore the shape of the loss landscape of those different architectures with various activation functions. Lastly, based on the result of our experimentation,
we introduce a new locally quadratic activation function namely Hytana alongside one variation Parametric Hytana which
outperforms common activation functions and addresses the dying ReLU problem.