Relevant Facial Key Parts and Feature Points for Emotion Recognition

Rim Afdhal, Ridha Ejbali and Mourad Zaied

Research Team on Intelligent Machines, National School of Engineers of Gab

es, University of Gab

es,

Avenue Omar Ibn El Khattab, Zrig Eddakhlania 6029, Gab

es, Tunisia

Keywords:

Emotion Recognition, Facial Expressions, Key Parts, Feature Points, Classiﬁcation Rates.

Abstract:

Interaction between people is more than just verbal communication. According to scientiﬁc researches, human

beings trust a lot on non-verbal techniques of communication, particularly communication and understanding

each other via facial expressions. Facial expressions are more descriptive in situations where words fail, such

as a surprise or a shock. In addition, lying via spoken words is harder to notice compared to faking expressions.

Focusing on geometric positions of facial key parts and well detecting them is the best strategy to boost the

classiﬁcation rates of emotion recognition systems. The goal of this paper is to ﬁnd the most relevant part

of human face which is responsible to express a given emotion using feature points and to deﬁne a primary

emotion by a minimum number of characteristic points. The proposed system contains four main parts: the

face detection, the points location, the information extraction, and ﬁnally the classiﬁcation.

1 INTRODUCTION

Emotions are basic characteristics of human beings

that play a relevant role in social communication

(Foggia.P et al., 2023). People use different man-

ners to express an emotion, like facial expression

(Tuncer.T et al., 2023), speech (Pawar and Kokate,

2021) (R.Ejbali et al., 2015) body language (Leong

et al., 2023) (I.Khalifa et al., 2019). The human

face, as the most expressive and exposed part of the

body (R.Afdhal et al., 2021), enables the use of com-

puter vision systems to analyze the images contain-

ing faces for emotions recognition. Facial expres-

sions (ELsayed.Y et al., 2023) have a relevant role

in recognizing emotions as well as are used in the

task of non-verbal communication, and to identify

humans. An automatic facial expressions analysis

is very relevant for automatic understanding of peo-

ple (Y.Siwar et al., 2016)(Y.Siwar et al., 2020), their

actions as well as their behavior. Facial expression

has been a hot topic of research in human behav-

ior (Huang.ZY et al., 2023). It ﬁgures surely in re-

search on almost every aspect of emotion such as

psychophysiology (Bourke C., 2010), neural corre-

late , perception, addiction, social processes, depres-

sion as well as other emotion disorders. Moreover,

it communicates physical pain (Rawal, 2022), alert-

ness, personality as well as well as interpersonal rela-

tions. Automated emotions detection using facial ex-

pressions is also fundamental for human–robot inter-

action (F. De la Torre and Cohn., 2015), drowsy driver

detection (Fuletra, 2013), analyzing mother–infant in-

teraction (A. R. Taheri and Basiri, 2014), autism, so-

cial robotics (So et al., 2018), as well as expression

mapping for video gaming.

This paper aims at presenting the following con-

tributions:

• Find the key part of the face which is responsible

to express a given emotion using feature points.

• Enhance a classic emotion recognition system by

decreasing the number of feature points in order

to recognize a basic emotion (joy, fear, disgust,

surprise, sadness and anger).

The remainder of this paper is structured as follows:

In section 2, we introduce literature related works.

Section 3 describes the proposed approach. Section 4

presents the results as well as related discussion. Fi-

nally, section 5 gives a conclusion of this work.

2 RELATED WORKS

In this section, we show the various works in the liter-

ature corresponding to our ﬁeld of interest. Two popu-

lar methods used mostly in the literature for the facial

expressions recognition systems are based on geom-

etry and appearance (Iqbal et al., 2023). The recog-

nition process of emotions runs up against the issue

1270

Afdhal, R., Ejbali, R. and Zaied, M.

Relevant Facial Key Parts and Feature Points for Emotion Recognition.

DOI: 10.5220/0012464300003636

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 16th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2024) - Volume 3, pages 1270-1277

ISBN: 978-989-758-680-4; ISSN: 2184-433X

of the complexity and the diversity of the human be-

ing expression and the relevant ﬂow of data which it

generates. The solution of this issue would be to de-

tect feature landmarks of the face and to extract for

them the useful data in the operation of recognizing

emotions.

The approach presented in (Tarnowski et al.,

2017) describes a model (Candid3) which is based on

121 characteristic landmarks of the face. These land-

marks are located on characteristic positions on the

face namely, the corners of the mouth, the nose, the

cheekbones, the eyebrows, etc.

(Salmam et al., 2018) have suggested an emotion

recognition technique using facial expression. The

technique was based on four facial key parts (eye-

brows, eyes, nose, and mouth). First, they detected

the face, then, they located and tracked 49 feature

points using a Supervised Decent Method. Then, they

calculate the distances between each pair of points.

Finally, to classify emotions they used a neural net-

work classiﬁer.

In this paper, we use an improved approach which

aims at decreasing the number of feature points and

boosting the classiﬁcation rates in order to recognize

an emotion.

3 PROPOSED APPROACH

In this section, we present an overview of our sug-

gested system for identifying primary emotions, fol-

lowed by a detailed description of the key components

of our proposed methodology. The ﬁgure 1 shows the

different steps of a classic emotion recognition sys-

tem. We enhanced this system by reducing the num-

ber of feature points at the level of location points

step.

3.1 Overview of Our System

In this paper, we proposed an improved approach aim-

ing at developing an emotion recognition system able

to recognize the six primary emotions with high clas-

siﬁcation rates and minimum number of characteristic

points.

The proposed system presents the importance of every

human face elements’ on the classiﬁcation phase or

the recognition of primary emotions. The system con-

tains four main steps. To detect the human face, we

used Viola-Jones detector. Then, we located the fea-

ture points. Besides, we extract information. Finally,

we classiﬁed the variations of biometric distances in

order to have the appropriate emotion using KNN (K-

Nearest Neighbor) and SVM classiﬁers.

We used two methods in order to acheive our goal.

The ﬁrst method is shown at ﬁgure2. We started our

work by a pre-treatment.

The different sub steps of the ﬁrst method are as fol-

lows:

• We divide the used data sets on subdatasets based

on the key parts of every emotion.

• We prepare the ﬁrst category of dataset which

based on three key parts (eyes,eyebrows and

mouth). We have two subdatasests (surprise and

disgust).

• We prepare the second category of dataset which

based on two key parts (eyes and mouth). We have

three subdatasests (sadness, fear and anger).

• We prepare the third category of dataset which

based on one key part (mouth). We have one sub-

datasest (joy).

• We detect the face using viola and jones detector.

• We locate the feature points on the face.

• We compute the variations of biometic disatnces.

• We use the KNN and the SVM for the classiﬁca-

tion.

The second method is shown at ﬁgure 3. The dif-

ferent sub steps of this method are as follows:

• We detect the face.

• We hide the down part of the face ( mouth and

nose).

• We hide the upper part of the face (eyes and eye-

brows).

• We locate the feature points on the not hidden

parts of the face.

• We compute the variations of biometic disatnces.

• We classify the obtained distances using SVM and

KNN.

3.2 Face Detection

The ﬁrst fundamental stage in all facial analysis tech-

niques is face detection (R.Afdhal et al., 2015). The

computer system needs some training to effectively

detect faces so that it can quickly determine if any-

thing is a face or not. Several threshold values have

been established in order to detect faces. A system

can identify human faces based on this value (Salmam

et al., 2018).

To complete this phase, we employed the Viola-Jones

detector (R.Afdhal et al., 2017). One of the most often

used techniques for object detection is the Viola-Jones

approach. This approach can offer results with high

Relevant Facial Key Parts and Feature Points for Emotion Recognition

1271

Emotion recognition system

Face detection

Input

Classified

Emotion

Output

Classification

Feature points

location

Information

extraction

Joy

Anger

Fear

Disgust

Surprise

Sadness

Viola and Jones

detector

Computing the variations

of biometric distances

SVM

KNN

Neural Networks

Location of 38 feature

points

D1 to D21

…

Figure 1: Classic emotion recognition system.

6 basic emotions

dataset

Method 1

Step1

Surprise

Sub dataset

Disgust

sub dataset

Sadness

Sub dataset

Fear

Sub dataset

Anger

Sub dataset

Joy Sub

dataset

Sub Step1: 3 key parts

Sub Step2: 2 keyparts -------------------------------------------------------------------------------------------------------------------------------------

Sub Step3: 1 key part -------------------------------------------------------------------------------------------------------------------------------------------

Step2: location of feature points based on key

parts

Sub Step4: face detection

Step3: feature extraction (computing the variation of biometric distances)

Exemple: Disgust Emotion

Step4: classification

𝒅

𝟏

𝒅

𝟐𝟏

𝒅

𝒆𝟏

𝒗𝒅

𝟐𝟏

𝒗𝒅

𝟏

𝒅

𝒆𝟐𝟏

Contribution:

Reducing

the

Number

Feature

Points

points

Training dataset

Dataset of test

SVM

KNN

Classified

emotion

Figure 2: The different steps of the ﬁrst method.

accuracy of about 93.7% with a speed of 15 times

faster. Viola-Jones detector uses rectangular Haar-

like features.

3.3 Feature Points Location

The feature points location is the second step of the

proposed system. We achieved this step using the ﬁve

rectangles technique (R.Afdhal et al., 2020). It’s a

very simple technique. It is based on 5 boxes detected

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

1272

6 basic emotions

dataset

Method 2

Step1: face detection

--------------------------------------------------------------------------------------------------------------------------------

Step2: location of feature points

Step3: feature extraction

Step4: classification

Training dataset

Dataset of test

Classification

Substep1

Substep2

𝒗𝒅

𝟐𝟏

𝒗𝒅

𝟏

𝒗𝒅

𝟐𝟏

𝒗𝒅

𝟏

Classified emotion

Dataset of test

Classification

Training dataset

Classified emotion

--------------------------------------------------

Figure 3: The different steps of the second method.

at the level of facial elements using Viola-Jones de-

tector. To precise the position of each landmark in

the human face, we focus on the result of the selec-

tion boxes (BBOX) of the face’s elements detection

(R.Afdhal et al., 2014). The ﬁrst box was located on

the face, the second on the right eye, the third on the

left eye, the fourth on the nose, ﬁnally the last box was

located on the mouth. The feature points were located

with the coordinates of the points characterizing each

box. The BBOX gives an M-by-4 matrix. It deﬁnes M

bounding boxes containing the detected objects. This

technique provides multi-scale object detection on the

input image. The output matrix row, BBOX, contains

a vector, [x y width height], that gives in pixels, the

upper left corner as well as the size of the box.

box

righteye

= [x

reye

width

reye

height

reye

] (1)

box

le fteye

= [x

leye

width

leye

height

leye

] (2)

box

f ace

= [x

f ace

width

f ace

height

f ace

] (3)

box

nose

= [x

nose

width

nose

height

nose

] (4)

box

mouth

= [x

mouth

width

mouth

height

mouth

] (5)

We will precise in this step the number of feature

points to every emotion. We used two methods as

mentioned above:

method1 : In this method, we are based on the

deﬁnitions of facial expressions with MPEG 4 Norm

and the facial key parts of every emotion mentioned

in (Plutchik, 1980). Table 1 presents the key parts of

each primary emotion and Table 2 shows the different

distances corresponding to each facial key part.Table

3 summarizes the distances as well as the feature

points of emotions.

Table 1: Key parts of primary emotions.

Emotion Key parts

Surprise Eyebrows, Eyes and Mouth

Disgust

Sadness Eyebrows and Mouth

Fear

Anger

Joy Mouth

Table 2: Key parts Distances.

Key part movements Distances

Eyebrow movements D1 to D7

Eye movements D8 and D9

Mouth movements D10 to D19

We will use for the emotion of surprise and dis-

gust three key parts eyebrows, eyes and mouth. So, to

recognize these two emotions we will use 34 feature

points and 19 biometric distances as presented at ﬁg-

ure 4.

We will locate for the emotion of sadness, fear and

anger two key parts eyebrows and the mouth. So, we

will eliminate the feature points located on the eyes.

To recognize these three emotions we will use 30 fea-

ture points and we will eliminate 4 points and 2 bio-

Relevant Facial Key Parts and Feature Points for Emotion Recognition

1273

Table 3: Distances and characteristic points of primary emotions.

Emotion Distances Characteristic points

Surprise Disgust D1 to D19 34

Sadness, Fear, Anger D1 to D7, D10 to D19 30

Joy D10 to D19 18

Figure 4: Static points- dynamic points - biometric dis-

tances of surprise and disgust emotions.

Figure 5: Characteristic points - biometric distances of sad-

ness, fear and anger emotions.

metric distances as presented at the ﬁgure 5. We will

use for the emotion of joy only the mouth so for this

emotion we will locate the feature points of the mouth

and we will eliminate all the points located on the eyes

and the eyebrows. So, we will use 18 feature points

and 10 biometric distances.

method2: The idea is to hide the key parts of the

face to each emotion in order to see the inﬂuence of

each part. Then, we computed the number of points to

locate for each emotion. We hid the lower part of the

face (nose and mouth). To classify emotions in this

case, we used only two key parts (eyes and eyebrows).

As well as, we located only the eyes and eyebrows

feature points. After that we hid the upper part of the

face (eyebrows and eyes). To classify emotions in this

case, we used only the mouth.

3.4 Information Extraction

An expressive human face is a deformation from the

neutral face. The differences between a neutral face

and its corresponding expressive face have lots of data

related to facial expressions. During the production

of a facial expression, a set of transformations in the

facial features appears on the face (R.Afdhal et al.,

2016). The idea is to classify different expressions

using the variation of some distances computed from

the coordinates of the points of interest that delimit

Figure 6: The biometric distances.

the eyes, nose, mouth and eyebrows from the neu-

tral state (R.Afdhal et al., 2020). For facial move-

ment processing, we represent each muscle by two

points, a static point and a dynamic point. Static

points are points that do not move during the pro-

duction of a facial expression, while dynamic points

undergo movement according to the type of expres-

sion. Figure 5 shows the different Euclidean distances

used to characterize the movement of the facial mus-

cles. D19 is the biometric distance that relates two

feature points p26 and p35. It is computed as follows:

D19 =

((p

(x) − p

(y))

+ (p

(x) − p

(y))

)

3.5 Classiﬁcation

In order to classify emotions, We choose the KNN

method. It is a simple Machine Learning algorithm

based on Supervised Learning technique, to catego-

rize emotions. The KNN technique places the new

instance in the category that is most similar to the

available categories by assuming similarity between

the new data and existing cases. A new data point is

classiﬁed using the KNN algorithm depend on simi-

larity after all the existing data has been stored. This

means that as fresh data is generated, it may be rapidly

and correctly categorized using the KNN technique.

These are the different steps of the algorithm:

• Choose the number K of the neighbors.

• Compute the Euclidean distance of K number of

neighbors.

• Take the K closest neighbors, determined by the

Euclidean distance.

• Compute the amount of data landmarks in each

category among these k neighbors.

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

1274

Table 4: Classiﬁcation rates of primary emotions of the Chon-kanade data base.

Emotion Method1 using SVM Method1 using KNN (L.Nwosu et al., 2017) (R.Afdhal et al., 2020)

Surprise 99% 98.5% 98.4% 99.3%

Disgust 97% 96% 97.2% 99.2%

Anger 94% 93% 93.7% 98.5%

Fear 80% 90% 92.2% 95%

Sadness 87% 85.5% 91.8% 92.31%

Joy 99% 98.5% 98.1% 98.5%

• Assign the additional data landmarks to the cate-

gory where the neighbor count is at its highest.

• The model is complete.

In order to enhance the emotions’ classiﬁcation rates,

we used the Support Vector Machine classiﬁer. It

is a preferment machine learning technique used for

nonlinear and linear classiﬁcation, regression, as well

as outlier detection issues. It can be used for many

applications, including face detection, spam detec-

tion, handwriting identiﬁcation, gene expression anal-

ysis, text classiﬁcation as well as picture classiﬁca-

tion. SVMs can handle high-dimensional data and

nonlinear interactions, making them ﬂexible in a wide

range of applications.

4 RESULTS AND DISCUSSION

This section shows the results obtained by the pro-

posed approach. As mentioned above, we used two

methods in order to precise the most relevant facial

key parts of every primary emotion. If an emotion is

deﬁned by a few number of key parts the number of

feature points will decrease.

We have used two datasets: the Chon-kanade and

JaFFE (Japanese Female Facial Expression) data ba-

sis. The Chon-kanade data basis has 593 images from

123 persons. The JaFFE data basis contains 213 im-

ages of different facial expression from 10 japanese

female persons.

Table 4 shows the classiﬁcation rates of our ﬁrst

method using SVM and KNN classiﬁers and the other

approaches (L.Nwosu et al., 2017) and (R.Afdhal

et al., 2020). This method (L.Nwosu et al., 2017)

is based on a two-channel convolutional neural net-

works. The extracted eyes is the input of the ﬁrst

convolutional layer and the mouth is the input of the

second convolutional layer. Information from the two

channels converge in fully connected layer which is

used to learn global information as well as is used for

classiﬁcation. The second method (R.Afdhal et al.,

2020) contains three phases: pre-treatment, feature

extraction as well as classiﬁcation. This method used

the deep learning for the feature extraction and the

fuzzy logic for the classiﬁcation. This comparison

is conducted to validate the performance of the sug-

gested method.

Table 5: Overall accuracy of JaFFE dataset.

Approach Classiﬁcation rates

(V.Chernykh et al., 2017) 73%

(Abate1 et al., 2022) 90,42%

(Chung et al., 2019) 72.16%

(Salmam et al., 2018) 93.8%

Our method 95%

As shown in table 5, obviously, our proposed

method outperforms the cited approaches by as much

as 22%, 4.58%, 22.84%, and 1.2% respectively.

The tables 6 and 7 present the results of the second

method. Table 6 presents the classiﬁcation rates when

we hid the lower part of the face (nose and mouth).

To classify emotions in this case, we used only two

key parts (eyes and eyebrows). As well as, we located

only the eyes and eyebrows feature points. The results

show the inﬂuence of the elimination of the feature

points located at the level of the mouth.

Table 6: The classiﬁcation rates using the upper part of the

face.

Emotion Rates with KNN Rates with SVM

Surprise 15% 25%

Disgust 10% 12%

Anger 60% 65%

Fear 50% 55%

Sadness 10% 13.5%

Joy 19% 20%

Table 7: The classiﬁcation rates using the down part of the

face.

Emotion Rates with KNN Rates with SVM

Surprise 45% 50%

Disgust 42% 55%

Anger 10% 15%

Fear 30% 35%

Sadness 35% 40%

Joy 80% 83%

As shown at table 6 the surprise and disgust were

classiﬁed with low classiﬁcation rates 15% with KNN

classiﬁer and 25% with SVM classiﬁer, 10% with

KNN classiﬁer and 12% with SVM classiﬁer, respec-

Relevant Facial Key Parts and Feature Points for Emotion Recognition

1275

tively. These low rates show the importance of the

mouth to classify these two emotions. We observed

also if we use the down part of the face the classi-

ﬁcation rates of these emotions increase. So, to per-

fectly classify surprise and disgust emotions, we must

use three facial key parts (mouth, eyebrows and eyes)

which conﬁrm our ﬁrst method. Anger, fear and sad-

ness were deﬁned according to our ﬁrst method by

two key parts (eyes and eyebrows).

Table 7 shows high classiﬁcation rates of anger

and fear emotions which prove the ﬁrst method. How-

ever, the sadness emotion was classiﬁed by low clas-

siﬁcation rates. So, for this emotion the mouth is rel-

evant to better classify this emotion.

Finally, joy emotion was deﬁned by one key part

(the mouth). The second method conﬁrms this deﬁni-

tion. When we hid the mouth, the emotion was classi-

ﬁed with low classiﬁcation rates(19% with KNN and

20% with SVM). However, table 7 shows high classi-

ﬁcation rates of this emotion when, we didn’t elimi-

nate the mouth.

5 CONCLUSION

The goal of this study is to deﬁne a primary emo-

tion (joy, fear, sadness...) by a minimum number of

characteristic points which will be very useful espe-

cially for real time applications. Our proposed system

consists of four phases: face detection, characteristic

points localization, information extraction and classi-

ﬁcation. We contribute to two levels: to assign the

most important facial key parts in order to perfectly

recognize a basic emotion and to reduce the num-

ber of characteristic points to deﬁne an emotion. The

rates in the experimental results present the effective-

ness of the suggested system.

ACKNOWLEDGEMENTS

The authors would like to acknowledge the ﬁnancial

support of this work by grants from General Direction

of Scientiﬁc Research (DGRST), Tunisia, under the

ARUB.

REFERENCES

A. R. Taheri, M. Alemi, A. M. H. R. P. and Basiri, N. M.

(2014). Social robots as assistants for autism therapy

in Iran: Research in progress. International Confer-

ence on Robotics and Mechatronics (ICRoM), Tehran,

Iran, 2nd edition.

Abate1, A. F., Cimmino, L., Narducci, B.-C. M. F., and

Pop, F. (2022). The limitations for expression recog-

nition in computer vision introduced by facial masks.

In Multimedia Tools and Applications.

Bourke C., Douglas K., P. R. (2010). Processing of facial

emotion expression in major depression: a review. In

Aust. N.Z. J. Psychiatry.

Chung, C.-C., Lin, W.-T., Zhang, R., Liang, K.-W., and

Chang, P.-C. (2019). Emotion estimation by joint fa-

cial expression and speech tonality using evolutionary

deep learning structures. In IEEE Global Conference

on Consumer Electronics (GCCE), pages 12–14.

ELsayed.Y, ELSayed.A, and Abdou.M.A. (2023). An au-

tomatic improved facial expression recognition for

masked faces. In Neural Computing and Applications,

page 14963–14972.

F. De la Torre, W. S. Chu, X. X. F. V. X. D. and Cohn., J.

(2015). IntraFace. IEEE International Conference and

Workshops on Automatic Face and Gesture Recogni-

tion, Ljubljana, Slovenia,, 11th edition.

Foggia.P, Greco.A, Saggese.A, and Vento.M” (2023).

Multi-task learning on the edge for effective gender,

age, ethnicity and emotion recognition. In Eng. Appl.

Artif. Intell.

Fuletra, J.D.; Bosamiya, D. (2013). Int. j. recent innov.

trends comput. commun. In Int J of Soc Robotics.

Huang.ZY, Chiang.CC, and Chen.JH (2023). A study on

computer vision for facial emotion recognition. In Sci

Rep.

I.Khalifa, R.Ejbali, and , M. (2019). Body gesture mod-

eling for psychology analysis in job interview based

on deep spatio-temporal approach. Parallel and Dis-

tributed Computing, Applications and Technologies:

19th International Conference, PDCAT, Jeju Island,

South Korea.

Iqbal, J. M., Kumar, M. S., G.R., G. M., A.N., S., Karthik,

A., and N., B. (2023). Facial emotion recognition

using geometrical features based deep learning tech-

niques. In International Journal of Computers Com-

munications and Control.

Leong, S., Tang, Y. M., Lai, C. H., and Lee., C. (2023).

Using a social robot to teach gestural recognition and

production in children with autism spectrum disor-

ders, disability and rehabilitation. In Assistive Tech-

nology.

L.Nwosu, H.Wang, J.Lu, I.U.X.Yang, and T.Zhang (2017).

Deep Convolutional Neural Network for Facial Ex-

pression Recognition Using Facial Parts. Int. Conf.

Dependable, Autonomic and Secure Computing, Or-

lando, FL, USA, 15th edition.

Pawar, M. and Kokate, R. (2021). Convolution neural

network based automatic speech emotion recognition

using mel-frequency cepstrum coefﬁcients. In Mul-

timed. Tools Appl. Springer.

Plutchik, R. (1980). A general psychoevolutionary theory

of emotion. In Theories of Emotion. Elsevier.

R.Afdhal, A.Bahar, R.Ejbali, and , M. (2015). Face de-

tection using beta wavelet ﬁlter and cascade classiﬁer

entrained with Adaboost. Eighth International Con-

ference on Machine Vision (ICMV), BARCELONA,

SPAIN.

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

1276

R.Afdhal, R.Ejbali, and , M. (2014). Emotion recognition

using features distances classiﬁed by wavelets net-

work and trained by fast wavelets transform. 14th In-

ternational Conference on Hybrid Intelligent Systems,

Kuwait.

R.Afdhal, R.Ejbali, and , M. (2016). Emotion recognition

using the shapes of the wrinkles. 19th International

Conference on Computer and Information Technology

(ICCIT), Bangladesh.

R.Afdhal, R.Ejbali, and , M. (2017). A Hybrid System

Based on Wrinkles Shapes and Biometric Distances

for Emotion Recognition. The Tenth International

Conference on Advances in Computer-Human Inter-

actions, Nice, France.

R.Afdhal, R.Ejbali, and , M. (2020). Emotion recognition

by a hybrid system based on the features of distances

and the shapes of the wrinkles. In The Computer Jour-

nal,.

R.Afdhal, R.Ejbali, and , M. (2021). Primary emotions

and recognition of their intensities. In The Computer

Journal,.

Rawal, N., S.-H. R. (2022). Facial emotion expressions in

human–robot interaction: A survey. In Int J of Soc

Robotics.

R.Ejbali, O.Jemai, M., and Amar, C. (2015). A speech

recognition system using fast learning algorithm and

beta wavelet network. 15th International Confer-

ence on Intelligent Systems Design and Applications

(ISDA), Marrakesh, Morocco.

Salmam, F. Z., Madani, A., and Kissi, M. (2018). Emotion

recognition from facial expression based on ﬁducial

points detection and using neural network. In Interna-

tional Journal of Electrical and Computer Engineer-

ing.

So, W.-C., Wong, M. K.-Y., Lam, C. K.-Y., Lam, W.-Y.,

Chui, A. T.-F., Lee, T.-L., Ng, H.-M., Chan, C.-H.,

and Fok, D. C.-W. (2018). Using a social robot to

teach gestural recognition and production in children

with autism spectrum disorders, disability and reha-

bilitation. In Assistive Technology, pages 527–539.

Tarnowski, P., Kołodziej, M., Majkowski, A., and Rak.,

R. J. (2017). Emotion recognition using facial expres-

sions. In International Conference on Computational

Science, pages 12–14, Zurich, Switzerland.

Tuncer.T, Dogan.S, and Subasi.A (2023). Automated fa-

cial expression recognition using novel textural trans-

formation. In J Ambient Intell Human Comput, page

9435–9449.

V.Chernykh, G.Sterling, and P.Prihodko (2017). Emotion

recognition from speech with recurrent neural net-

works.

Y.Siwar, J.Olfa, E.Ridha, and , Z. (2016). Improving the

classiﬁcation of emotions by studying facial feature.

International Journal of Computer Theory and Engi-

neering.

Y.Siwar, S.Salwa, and , Z. (2020). A novel classiﬁcation

approach based on Extreme Learning Machine and

Wavelet Neural Networks. Multimedia Tools and Ap-

plications.

Relevant Facial Key Parts and Feature Points for Emotion Recognition

1277