Professional Documents
Culture Documents
My Abstract
My Abstract
Architecture
Aiman Shahid, Hamad Rizwan
Department of Computer Science Department of Computer Science,
University of Engineering and University of Engineering and
Technology Technology
Taxila, Pakistan Taxila, Pakistan
18-cs-21@students.uettaxila.edu.pk 18-cs-40@students.uettaxila.edu.pk
Abstract— Emotions are a natural and significant visual processing sciences to affective computing
element of human behavior that influence how we and computer animations [2].
communicate. For a greater understanding of human
behavior, precise analysis, and interpretation of the Automatic facial emotion recognition (FER) (An
emotional content of human facial expressions are expanded variant of the acronym FER is used in
required. AI has gained popularity in almost every each study, such as facial emotion recognition and
field. Deep learning methods have been investigated as
a series of strategies for achieving robustness and facial expression recognition.) is a popular topic.
scalability on new types of data. Deep learning is used Because this study focuses on the broad
in this article to recognize human emotions through characteristics of facial emotion recognition, the
facial expressions. We have built a novel CNN model word FER refers to facial emotion recognition in
that can predict human expression from an image.
this work.) With the fast growth of artificial
CK+ and FER2013 dataset was used to experiment
with and train a deep convolutional network in this intelligence approaches, such as human-computer
case. We also applied augmentation techniques like the interface (HCI) [3,4], virtual reality (VR) [5,6],
K-fold method while working with the CK+ dataset. augmented reality (AR) [6, advanced driver
Furthermore, we attained an accuracy rate of 89 assistant systems (ADASs) [7], and entertainment
percent with CK+ and 69 percent with the FER2013
dataset which was far better than many systems. Using [8,9], has also been expanding recently. Although
our CNN model, we recognized six different emotions. many sensors for FER inputs such as
When tested with different people under varying electromyography (EMG), electrocardiogram
ambient and light conditions, our work has shown
positive outcomes. Results were evaluated using the
(ECG), electroencephalograph (EEG), and camera
confusion matrix and roc curve. may be used, a camera has been the most successful
sensor that detects since it gives the most useful
Keywords— facial emotion recognition, deep learning- information for FER and does not require
based facial emotion recognition, classification, CNN, k- maintenance.
fold technique.
In this paper, we present a deep learning-based
I. INTRODUCTION emotion detection model based on facial
expressions from photographs. Unlike previous
Facial expressions are key components of human
applications, the system maintains high accuracy
communication because they help us perceive
outcomes in a variety of environments and models.
others' intentions. People use facial expressions and
In our experiment, we developed a novel CNN
verbal tones to infer other people's emotional states,
model. In our experiment, we used two datasets:
such as joy, sadness, and aggression. Verbal
FER2013 and CK+. The CNN model was then
components transmit one-third of human
trained using both datasets. Furthermore, the
communication, whereas nonverbal components
accuracy of the produced CNN's was checked by
convey two-thirds, according to different research
comparing them to test data.
[1,2]. Facial expressions are one of the most
important nonverbal components in social
communication because they contain emotional II. LITERATURE REVIEW
significance. As a result, it's not a bit surprising that Deep learning has addressed issues in a variety of
face emotion research has gotten a lot of attention fields [9], while previous techniques involved
in recent decades, with applications ranging from manually gathering data and making decisions
based on a set of rules. The sensory mistake is
highly likely to occur in the earlier approaches, and recognize a face, facial expression, eye, and head
decisions are made based on a human-built set of movement to determine the learner's interest,
rules, which may or may not be always right. boredom, and so on during learning. In comparison to
Researchers have used a variety of approaches to other modern systems, machine learning and deep
classify human face expression, including support learning algorithms be more efficient in terms of
vector machines [10], principal component analysis implementation and outcomes [21].
[11], and convolutional neural networks [12, 13]. CNN can not only recognize the face and facial
The majority of real-time emotion detection expressions, but it can also be fine-tuned to detect
research has focused on visuals. certain areas of the face rather than the entire face.
Furthermore, on numerous datasets, including FER-
When the studies in the literature on facial emotion
2013, CK+, FERG, and JAFFE [22] [23] [24], the
identification with computer vision techniques are
CNN model outperforms previous approaches.
evaluated over the last several years, they indicate a
Even though there has been a lot of work done on
considerable increase in the development of
emotion recognition, it has all been focused on
computer technology.
discovering the fundamental emotion, but the
Even though there are several applications for facial complex emotions that humans confront during the
recognition and detection. Breuer and Kimmel [14] learning process have been neglected.
employed visual approaches to interpret a CNN
model for emotion identification that was trained
with a range of datasets. They put CNN to the test III. PROBLEM STATEMENT
on datasets for face emotion detection as well as In the fields of computer vision and artificial
several facial emotion recognition applications. intelligence, facial emotion recognition plays a
Jung et al. [15] devised a method that combined critical role. Emotion is one of the fundamental
two forms of CNNs. One of them is the expressions of human beings. It is an integral part
identification of geometric aspects of facial parts, of our nonverbal communication. It helps in
while the other is feature extraction from visual determining a person's ideas, behavior, and feelings.
data. Based on facial expressions, we provide the
Liu et colleagues [16] categorize distinct facial findings of recognizing five emotional states
expressions using the FER2013 dataset and a two- (neutral, Happy, sad, anger, and fear). To predict
layer CNN. They also compared it to four other the grayscale picture of a person’s face, which
existing models, finding that the suggested model had emotion the facial expression conveys. Our
a test accuracy of 49.8%. S Suresh et al [17] created a evaluation metric will be the accuracy for each
Deep Neural Network-based sign language emotion (fraction of correctly classified images)
recognition system that distinguishes six distinct sign [12], supplemented by confusion matrix which
languages (DNN). When two models with different highlights which emotions are better recognized
optimizers (Adam and SGD) are compared, it is than others. Our system would be to understand a
discovered that the model with the Adam optimizer is face and its characteristics and then make a
more accurate. K. Bouaziz et al [18] have exhibited weighted assumption of the identity of the person.
an analytics approach that incorporates picture Input: 48 by 48 grayscale image of a face
recognition tools and methodologies. The suggested
Output: Emotion conveyed by facial
model uses CNN architecture to classify different
expression
types of handwriting. Automatic facial expression
recognition was achieved in [19] by extracting
features using wavelets modification and classifying
emotions using the K-nearest neighbors (KNN)
algorithm.