An Efficient Technique for Global Facial Recognition using Python and

OpenCV in 2D Images

JOSÉ CADENA, MANUEL VILLA, MAIRA MARTÍNEZ, JAIME ACURIO, LUIS CHACÓN

Universidad Técnica de Cotopaxi,

Facultad de Ciencias de la Ingeniería y Aplicadas,

Av. Simón Rodríguez, Latacunga

ECUADOR

Abstract: - The present work is an investigation that deals with the use of efficient techniques for global facial

recognition using Python and OpenCV carried out in the Information Systems career of the Faculty of

Engineering and Applied Sciences of the Technical University of Cotopaxi. We work with a database of 2D

faces corresponding to the students of the Information Systems career that served for the analysis and

comparison of the three techniques used (Fisherfaces, EigenFaces, LBPH). The objective of our work is to

determine an efficient technique that contributes to the area of global facial recognition, contributing

significantly to the field of university security, taking into consideration the saving of resources for future

implementations. Finishing this work with the respective analysis and interpretation of the research results, it

has been determined that of the three techniques studied, LBPH has the best results both in training time and in

face recognition efficiency, reaching near 100% in our tests.

Key-Words: - Facial recognition, OpenCV, Python, Fisherfaces, LBPH.

Received: October 9, 2022. Revised: January 13, 2023. Accepted: February 15, 2023. Published: March 23, 2023.

1 Introduction

This research project describes the facial recognition

techniques used, in addition to working with Python

and OpenCV, the same ones that apply artificial

intelligence algorithms in 2D images.

The objective of this work is to find an efficient

technique for a 2D facial recognition process and to

test its effectiveness through a prototype designed

with this technique.

Face recognition systems are a problem that is

still the subject of research because a large part of

the problem is due to factors that can affect the

system at the time of recognition.

According to [1], some factors hinder the facial

recognition process, such as occlusions that prevent

good recognition, whether due to gestures made by

the person, elements that cover the face, lighting,

distance, etc. scars, and aging among other factors.

The Windows operating system was installed

with the Python programming language, the

OpenCV, os, and numpy libraries, and the PyCharm

IDE in which the face recognition algorithm was

developed using the technique that provides the best

result to determine an efficient technique.

For [1], to check the image processing capacity

for facial recognition, three techniques were used:

Eigenfaces, LBPH (Local Binary Patterns

Histograms), and Fisherfaces.

The problem of facial recognition is very

complex since a person's face can vary depending

on their age, gestures, emotions, half-covered faces,

light intensity, etc., this has insisted that the process

of recognition does not reach 100% efficiency.

At present many techniques allow these facial

recognition processes to be carried out, they try to

reach a high percentage of effectiveness, with this

work a very acceptable level was reached, through a

bibliographic review of databases, feature extraction

techniques, and pattern classification algorithms in

Python and OpenCV.

Finally, to check if the applied technique meets

the expected expectations, different field tests will

be carried out with the detection of faces and the

comparison of results in the three techniques

described.

2 Problem Formulation

Nowadays, the use of facial recognition has been

one of the most common problems worldwide due

to the availability of software, the cost, and the

hardware needed for its implementation in a public

or private organization. World powers such as the

United States, Russia, China, Japan, Germany, and

WSEAS TRANSACTIONS on SYSTEMS and CONTROL

DOI: 10.37394/23203.2023.18.5

José Cadena, Manuel Villa,

Maira Martínez, Jaime Acurio, Luis Chacón

E-ISSN: 2224-2856

Volume 18, 2023

North Korea allocate large amounts of economic

resources to the implementation of facial

recognition tools in companies in these countries so

that they have greater security in the authentication

of workers.

According to [2], due to the current situation of

violence and crime that Ecuador is going through, it

is necessary to propose solutions that reduce

insecurity rates, a facial recognition platform was

implemented, designed to be exposed to the public

through services web, the logic deployed was

structured to satisfy the requirements in the context

of the country, seeking integration with the entities

in charge of citizen security by having control of the

development of the platform, different techniques

and changes to the face detection algorithms were

combined. and face recognition.

According to [3], in primitive civilizations

people lived in small communities, where they were

easily recognized by their limited number of

inhabitants; however, due to the rapid population

increase and human mobility, identification has

become a complicated process, in such a way that

innovative societies have found it necessary to

implement sophisticated identification recording

techniques.

[4], stated, in 1882, the French policeman

Alphonse Bertillon (1853-1914) presented the first

system of identification of people based on physical

characteristics, that is, on biometric features, and

called anthropometry and this is considered the first

scientific system used by the police to identify

criminals, it works by classifying the shape of the

nose, face or body of people: The following image

shows the different types of noses that were

published in Pearson's Magazine. It is presented in

Figure 1.

Fig. 1: Image published by Pearson's Magazine [4].

[5], affirms that the concept of facial detection

began in the '60s. Between the years 1964 and 1965

Woodrow Wilson, Helen Chan, and Charles Bisson

developed the first semi-automatic facial

recognition system through the use of the computer,

in the 70s Goldstein, Harmon, & Lesk, used 21

physical characteristics among them the color of the

hair, the thickness of the lips, etc. To improve facial

detection, and to identify these features, a manual

sequence had to be followed.

In the early 1990s, [6], using the "eigenfaces

technique" (method discovered by Kirby and

Sirovich) demonstrated that: "the error was used in

the identification of faces in images, this was a

finding that allowed perform face detection systems

in real-time. That is why the similarity was forced

by environmental factors, however, this caused

significant interest in the further development of

these systems.

[7], stated in 2001, surveillance cameras were

used in a Super Bowl game where the procedure

was the collection of images through the cameras to

later be compared with digitized images of

delinquents who were in a database that stored said

information.

For [8], Machine Learning is a technique

associated with the automatic detection of relevant

patterns within a data set. In recent years, it has

become a very common tool in practically all tasks

that require extracting information from data. From

large amounts of data.

According to [9], on a day-to-day basis, we are

surrounded by technology based on Machine

Learning: email filtering, recommendation systems,

facial detection, smartphone speech recognition,

weather forecasting, etc., or consulting traffic on the

road, it is also used in other fields such as medicine,

marketing, logistics or the maintenance of industrial

equipment.

[9] state, due to the complexity of all these

applications, a human being is not capable of

programming a series of specific specifications to

carry out said tasks, but rather has to provide the

computers themselves with the ability to learn from

experience and adapt to new situations.

According to [10], learning refers, as has been

seen, to a wide spectrum of situations in which the

learner increases his knowledge or his abilities to

accomplish a task, learning applies inferences to

certain information to build an appropriate

representation. Of some relevant aspect of reality or

some process, a common metaphor in the area of

machine learning within Artificial Intelligence is to

consider problem-solving, according to [11], as a

type of learning that consists once a type of problem

has been solved. A problem in being able to

recognize the problematic situation and react using

the learned strategy, currently the greatest

distinction that can be drawn between an animal and

a problem-solving mechanism is that certain animals

are capable of improving their performance, in a

broad set of tasks, as a result of having solved a

certain problem.

WSEAS TRANSACTIONS on SYSTEMS and CONTROL

DOI: 10.37394/23203.2023.18.5

José Cadena, Manuel Villa,

Maira Martínez, Jaime Acurio, Luis Chacón

E-ISSN: 2224-2856

Volume 18, 2023

For [11], computer vision, also called artificial

vision, machine vision, computational vision, image

analysis, or scene interpretation, is the process of

extracting information from the real world from

images using a computer as a tool.

According to [12], face detection is a technique

that allows finding the face of one or more people in

an image, while ignoring the background of the

image or other objects that are present within it.

According to [13] is present in many

applications that we use every day, for example on

Facebook the images that we upload already detect

the face, and what often asks us is to add the name

of the person or persons that appear there. , we also

have Instagram filters where the face needs to be

detected so that the filters that we want to apply are

used. It is even present in some banking applications

or when we use a smartphone to take a photo and it

shows the faces of the people who appear there in a

box or circle.

For [12], object detection using cascade

classifiers based on de Haar functions is an effective

object detection method proposed by Paul Viola and

Michael Jones, it is an approach based on machine

learning in which a cascade function is trained on

many positive and negative images, then used to

detect objects in other images. The method used by

Haar can be described as a screen of pixels, of

different orientations and sizes, separated by

rectangles, each rectangle being either positive or

negative.

According to [14], facial recognition is an area

that is part of pattern recognition, in recent years it

has gained great interest, especially due to the wide

range of applications it has in different fields such

as security, surveillance, cards, smart, among and

others.

On the other hand, [15], facial recognition is a

part of computer vision. Facial recognition has been

used for many decades, mainly by the armed forces.

According to [16], it is understood that facial

recognition is a part of computer vision, it is

necessary to know its definition to have a better

understanding of the subject, it is a way to obtain,

and process images of the real world to obtain

numerical information that can be handled by a

computer.

For [7], facial recognition is based on patterns

that can be checked if an image contains a face. This

is achieved by training the neural network methods

with images that contain faces and others that do not

show images.

For [17], facial detection systems aim to detect

the presence of a person through their facial features

in a digital image, the main advantage of these

detection systems is that it is not intrusive, so it does

not require the collaboration of the user beyond

being in front of the camera used in the security

system, in addition to the fact that it only requires a

single capture device.

For [18], the local binary pattern method was

designed for the description of textures. According

to [19], the use of local descriptions in some regions

of the face provides more information than others,

so texture descriptors tend to average the

information they describe, which is not convenient

when describing faces since maintaining

information on spatial relationships is important.

For [20], to form the global description, the image

of the face is divided into different regions, to which

a histogram is applied with which the LBPH

operator is obtained, which describes independent

information by region, these local descriptions are

then concatenated. To build a global description of

the face.

[21], Fisherface is a face recognition technique,

which takes into account light and facial

expressions. This is in charge of classifying and

reducing the dimensions of the faces using the FLD

(Discriminant Linear Fisher) method. For [21],

Fisher's discriminant analysis tries to project the

data in such a way that its new dispersion is optimal

for classification, while PCA looks for the vectors

that best describe the data, LDA (Discriminant

Linear Analysis) looks for the vectors that provide

better discrimination between classes after the

screening.

According to [22], Fisherfaces performs an

LDA, where it seeks to take advantage of the

available information about the classification of the

training images, to find a projection that maximizes

the separation between images of different people

(or classes) and minimizes the distance between

images of the same class, thus concentrating the

images, significantly improving the recognition rate.

Several techniques are used for facial recognition,

but these have shortcomings when implemented in

some companies such as response time and analysis

of 2D image patterns, generating a loss of time at

the time of user authentication and thus damaging

the arrest of the faces.

[23], suggests that Python is a high-level

language since it contains some implicit data

structures such as lists, dictionaries, sets, and tuples,

which allow performing some complex tasks in a

few lines of code and in a readable way.

The Python Standard Library, [24] states the

precise syntax and semantics of the Python

language, this library reference manual describes the

standard library that ships with Python, and it also

WSEAS TRANSACTIONS on SYSTEMS and CONTROL

DOI: 10.37394/23203.2023.18.5

José Cadena, Manuel Villa,

Maira Martínez, Jaime Acurio, Luis Chacón

E-ISSN: 2224-2856

Volume 18, 2023

describes some optional components that are usually

included in Python distributions.

3 Problem Solution

The focus of this research is quantitative, therefore,

it was possible to store and analyze the data

obtained in a way that facilitates the research, to

better understand the object of study on global facial

recognition using Python and OpenCV, with this

information raised the problem, formulation of the

problem, theoretical foundation, processing and

discussion of the results.

To determine the sample, three facial

recognition techniques were used: Eigenfaces,

Fisherfaces, and LBPH with a total of 31 students of

different characteristics and ethnicities will be used

in this research, students from the Information

Systems career were used to conduct research,

obtaining data and therefore the results of the

research.

This research will handle 2D facial images.

Facial images have unique characteristics that

differentiate them from one another. A human face

is established in two dimensions by taking the plane

and coordinates (x, y) in such a way that we can

obtain images, with gestures and emotions (happy,

sad, angry, etc.). It is presented in Figure 2.

Fig. 2: Emotions and gestures.

As part of the experimental research applied in this

research work, for the analysis and processing of the

data obtained, algorithms are used to extract

pickpockets from images and algorithms for their

recognition within the 2D recognition processes.

Face detection is a technique that makes it

possible to find the face or faces of numerous

people in an image while ignoring the background

of the image or other objects that are present in it.

It is present in many applications that we use

every day, for example, in the social network

Facebook with the images that are uploaded detect

the face, and what it often asks us is to add the name

of the person or persons that appear there. Also in

the social network Instagram, we have filters where

the face needs to be detected to be applied.

3.1 Detecting a Face in an Image

It is not an easy task for the computer, so we need it

to 'learn', which is why we use Machine Learning or

automatic learning for face detection.

Initially, a large number of images are needed to

train a classifier, so that it can differentiate between

the presence of an object and its non-presence. For

example, to perform a face detector, positive images

(containing faces) and negative images (images

containing no faces) are required.

Then we will proceed to extract features from

all images, to then use a Machine Learning

approach, and proceed with training. Finally, the

classifier can be obtained. It is presented in Figure

Fig. 3: Process to create a face classifier, [34].

Now that we have this process in mind, let's briefly

look at facial detection using the Haar Cascade. This

object detector uses the method proposed in the

paper “Rapid Object Detection using a Boosted. It is

presented in Figure 4.

WSEAS TRANSACTIONS on SYSTEMS and CONTROL

DOI: 10.37394/23203.2023.18.5

José Cadena, Manuel Villa,

Maira Martínez, Jaime Acurio, Luis Chacón

E-ISSN: 2224-2856

Volume 18, 2023

Fig. 4: Process to create a face classifier using Haar-

Cascade, [34].

3.2 Haar Cascades Python – OpenCV

OpenCV offers us pre-trained classifiers not only

for people's faces, which we will use in this work

but also for eyes, smiles, and whole bodies, among

others. You can find these XML files in the folder:

OpenCV/data/haarcascades/ or in the OpenCV

repository on GitHub.

3.3 DetectMultiScale

To use face detection with haar cascade in OpenCV

we are going to need the DetectMultiScale module

that will help detect objects according to the

classifier used. This will allow us to obtain a

delimiting rectangle where the object to be found

within an image is located, for this, we must specify

some arguments that are detailed below.

3.4 Scale Factor

This parameter specifies how much the image will

be reduced. For example, if 1.1 is entered, it means

that the image will be reduced by 10%, with 1.3 it

will be reduced by 30%, thus creating a pyramid of

images. It must be taken into account and that is

that, if we give a very high number, some detections

are lost. While for very small values such as 1.01

(that is, to reduce the image by 1%), it will take

longer to process, since there will be more images to

analyze, in addition to the fact that they can increase

false positives (which are detections presented as

objects or faces, but in reality, are not). It is

presented in Figure 5.

Fig. 5: Example of an image pyramid, [24].

An image pyramid is made due to the size of the

faces in the image, some may occupy more or less

area than others so to try to detect all in their

different sizes, the image pyramid is applied.

3.5 MinNeighbors

This parameter specifies how many neighbors each

candidate rectangle must have to retain it. We have

a small window that will go through an image

looking for faces, so you may find that at the end of

the whole process, it has identified several faces

(Figure 7), but many of them may correspond to the

same person. So, this parameter relates to all those

delimiting rectangles of the same face. Therefore,

minNeighbors specifies the minimum number of

bounding boxes, or neighbors, that a face must have

to be detected as such. It is presented in Figure 6.

Fig. 6: Face detection.

WSEAS TRANSACTIONS on SYSTEMS and CONTROL

DOI: 10.37394/23203.2023.18.5

José Cadena, Manuel Villa,

Maira Martínez, Jaime Acurio, Luis Chacón

E-ISSN: 2224-2856

Volume 18, 2023

Now, we must take into account that the higher the

value that we set, the fewer faces will be detected,

while if a very low value is given, false positives

can occur.

3.6. Detection of Faces on an Image

To begin with, the XML file containing the frontal

face classifier

(haarcascade_frontalface_default.xml) is saved in

the same folder where the script is stored. We are

going to detect the faces that appear in the following

image. The process is presented in Figure 7, Figure

8, and Figure 9.

Fig. 7: Input image for face detection.

The programming given for face detection will be as

follows:

Fig. 8: Code to detect the face.

Line 1: Import OpenCV.

Line 3: Load the classifier with an XML extension

with the help of cv2.CascadeClassifier, inside

quotes specify the name and extension of the file.

Lines 5 and 6: Read the image where the faces will

be detected and then proceed to transform it to

grayscale.

Line 8: Load the classifier in line 3 it is necessary to

apply it to the image, for this purpose

detectMultiScale is used, which must be followed

by the variable with which the load of the classifier

was assigned and this is where the recently raised

arguments are: image, scaleFactor, minNeighbors,

minSize, and maxSize.

Fig. 9. Image with face stop.

3.7 Storing Faces

We proceed to reuse the face detection code in an

image with the use of haarcascades, we will use the

same procedure adding new lines of code to save the

faces. It is presented in Figure 10.

Fig. 10: Input image to store faces.

The code to detect and store the faces of the image

will be the following (Figure 11).

Fig. 11: Code to store faces from an image.

WSEAS TRANSACTIONS on SYSTEMS and CONTROL

DOI: 10.37394/23203.2023.18.5

José Cadena, Manuel Villa,

Maira Martínez, Jaime Acurio, Luis Chacón

E-ISSN: 2224-2856

Volume 18, 2023

Line 1: We import OpenCV.

Line 3: Load the face classifier.

Line 5: We proceed to read the input image that

would correspond to the image in Figure 12.

Line 6: Create a copy of the input image, this will

remain intact, so it will serve to govern the faces

found.

Line 7: Transform the input image from BGR to

grayscale.

Line 9: The detection of faces present in the image

is performed, according to the input parameters of

the detectMultiScale function.

Line 11: Start a counter count = 0, which helps us to

count each of the detected faces.

Line 13: Extract the coordinates, width, and height

of each of the detected faces.

Line 15: Display the detected faces in a pink

rectangle.

Line 16 and 17: In line 16 we proceed to crop the

faces of the imageAux image with the help of the x,

y width, and height coordinates that we obtained

from line 13. In line 17 the faces are resized to 150

pixels wide and high.

Line 18: Each of the images will have a different

number thanks to the count counter.

Lines 21 to 25: These lines are for displaying the

face and detections.

Running the code returns the following as presented

in Figure 12.

Fig. 12: Storing faces from an image.

As any key is pressed, a face is surrounded and

stored in the folder where the script is located. It

should be taken into consideration that all the faces

have the same width and height since we resized

them in the code.

3.8 Store Faces Present in a Video

For the extraction and storage of faces through the

use of videos, it can be done in a streaming video or

read a video. For example, we will do it in a

streaming video as presented in Figure 13.

Fig. 13: Code to store faces from a video

Line 6: Specify that a video will be streamed,

however, a video can be read. Cv2.CAP_DSHOW is

used so that at the time of visualization they

completely occupy the window.

Running the code returns the following. They are

presented in Figure 14, and Figure 15.

Fig. 14: Storing faces.

Fig. 15: Stored Faces

This procedure will be used later for facial

recognition since we need a large number of images

of people to be recognized by the algorithm.

To carry out facial recognition, in the first place

it is necessary to collect data, that is, the faces of the

people that you want to recognize, then proceed to

train the classifier, to finally test it. For all this

process it will be necessary to use face detection

with haarcascades.

3.9 Creation of the Database

To perform facial recognition, the faces of the

people who want to recognize themselves are

necessary. These faces must denote a variety of

expressions such as: happiness, sadness, boredom,

and surprise, among others. Another aspect that

these images must have is the variation in light

conditions, whether people wear glasses or not, even

whether they close their eyes or wink.

It is recommended that these images be

collected in the setting or environment where facial

WSEAS TRANSACTIONS on SYSTEMS and CONTROL

DOI: 10.37394/23203.2023.18.5

José Cadena, Manuel Villa,

Maira Martínez, Jaime Acurio, Luis Chacón

E-ISSN: 2224-2856

Volume 18, 2023

recognition is to be applied. All this variety of

images obtained from the faces will contribute to the

performance of the algorithms used in this work.

In the code used for the demo, 130 faces from a

streaming video will be stored automatically. It is

presented in Figure 16.

Fig. 16: Code to store faces in the database.

Line 1 to 3: Import OpenCV, os and imutils.

Lines 5 to 7: In these lines, a folder is created with

the name of the person you want to recognize, this

will be created inside the Data folder that you had

previously created manually. So in line 5 in

personName, the name of the person is assigned,

and in line 6 dataPath is assigned the location of the

directory where each folder will be created with the

name of each person to be recognized. Finally,

personPath will be the full path.

Lines 9 to 11: With the information from the last

lines, the directory will be created with the name of

the person to be recognized inside the Data folder.

In line 5, for example, you can see that a folder

called 'Tutillo_9no_Sis1' will be created as

presented in Figure 17.

Fig. 17: Code to establish the number of images.

Line 23: Resize with imutils.resize, this is done to

resize the input video frames.

Line 32: The images corresponding to the faces are

resized so that they all have the same size. 150

pixels have been set.

Line 38: In this line, I have added the condition of

count >= 130 so that the storage process ends at 130

stored faces. Running the code returns the

following. It is presented in Figure 18.

Fig. 18: Database created from the established code.

3.10 Preparing Data for Training

To proceed with the training, it is necessary to have

each of the images with their respective label

associated with the person to whom the faces

belong. For example, when we read the folder

'Abata, all those images will be assigned a label 0,

then all the images of the faces of 'Caizaluisa' will

be assigned 1, 'Canizares' will be assigned 2, and so

on. With each of these labels, the computer will

know that the images correspond to different people.

In another class that we call 'RFtrainer.py' this

process is prepared, that is, the images and labels to

train them used: Eigenfaces, fisherfaces, and Local

Binary Patterns Histograms. It is presented in Figure

19.

Fig. 19: Code to prepare the images for training.

Line 5 to 7: First, specify the path of the 'Data'

folder that was created manually before and within

it will be each folder with the name of the people

you want to identify. Line 6 lists all the names of the

folders stored in 'Data'. Line 7 prints the list

obtained.

WSEAS TRANSACTIONS on SYSTEMS and CONTROL

DOI: 10.37394/23203.2023.18.5

José Cadena, Manuel Villa,

Maira Martínez, Jaime Acurio, Luis Chacón

E-ISSN: 2224-2856

Volume 18, 2023

Line 9: Labels are declared in this the labels

corresponding to each image will be stored

according to the person.

Line 10: faces Data is declared where each of the

face images will be stored.

Line 11: A label counter is set to 0, so as it finishes

reading the images of a person, it changes to another

value. This will help the classifier understand what

it takes from different people.

Line 13 to 15: Read each folder inside 'Data'. The

person Path sets the path to each person's folder.

While in line 15 we print the message 'Reading the

images...' so that when it is executed it can know in

which part of the process it is.

Lines 17 and 18: All the images corresponding to

each face are read. On line 18 the name of the folder

and the image are printed.

Line 19: In labels, the labels of each image are

added.

Line 20: Each image (face) is added to the faces

Data array.

Line 24: Every time the faces and labels of a folder

are finished storing, the label will be increased by 1.

3.11 Training

Starting from facial recognition using OpenCV, we

move on to code. They are presented in Figure 20,

and Figure 21.

Fig. 20: Methods for training the knower.

Fig. 21: Training the face recognizer.

Line 39: To train the face recognizer,

face_recognizer.train is needed, where

face_recognizer will be the variable where the

method was assigned (any of lines 31, 32, or 33).

Within the parentheses, you have to specify the

array where the faces or training images are

contained, while the second parameter corresponds

to the labels. It is necessary that this be a numpy

array so I have put np.array(labels).

3.12 Save the Obtained Model

Once the model that helps us to recognize faces has

been trained, we proceed to save it and read it in

another class. To store the model, write is used,

within the parentheses the name to be assigned is

specified, together with the XML or YAML

extension. It is presented in Figure 22.

Fig. 22: Save obtained model

Running the code returns the following. They are

presented in Figure 23, and Figure 24.

Fig. 23: Training the face recognizer.

Fig. 24: Saving the obtained model.

4 Results

4.1 Efficiency Comparison between Facial

Recognition Techniques

The success of the EIGENFACES technique

corresponds to 90% effectiveness and 10%

inefficiency, identifying 28 hits and 3 refusals out of

a total of 31 students with 4030 photographs of

which 3100 were used for training and 930 for the

same tests. They have features like: gestures, light

variations, and props

The success of the FISHERFACES technique

corresponds to 100% effectiveness, identifying 31

WSEAS TRANSACTIONS on SYSTEMS and CONTROL

DOI: 10.37394/23203.2023.18.5

José Cadena, Manuel Villa,

Maira Martínez, Jaime Acurio, Luis Chacón

E-ISSN: 2224-2856

Volume 18, 2023

correct answers out of a total of 31 students with

4030 photographs of which 3100 were used for

training and 930 for tests, which have characteristics

such as: gestures, variations of light and accessories.

The success of the LBPH technique corresponds to

100% effectiveness, identifying 31 successes out of

a total of 31 students with 4030 photographs of

which 3100 were used for training and 930 for the

tests, which have characteristics such as: gestures,

variations of light, and accessories. It is presented in

Table 1.

Table 1. Times and hits obtained in the test phase

In summary, according to Table 1, we worked with

4030 student photographs, 130 photographs per

student. Of which 100 photographs were taken for

training and 30 for tests. That is 3100 images for

training and 930 for tests.

As a result, the FISHERFACES and LBPH

techniques are more precise in terms of success in

recognition. Based on the time spent and success, it

is concluded that the LBPH is the most efficient

technique.

4.2 Technique Recognition Average Time

Once the success time of each student for each

technique was obtained (see Table 2), a comparison

was made between the times of the techniques, in

order to identify the efficiency in the response time.

The LBPH technique with an identification average

of 1.64 seconds.

Table 2. Statistics on the time spent in the facial

recognition process.

5 Conclusion

The bibliographic review of several books, scientific

articles, and web pages related to the 2D image

database, feature extraction, and pattern recognition

algorithms was of vital importance for the

development of the theoretical foundation and the

determination of technological tools. With the

design of the prototype for global facial recognition

using Python and OpenCV, he allowed the

comparison of the techniques used to determine

which technique is the most efficient. Through the

endorsement of expert judgment, it was determined

that the prototype for facial recognition complies

with the correct operation for face detection and in

an analysis of the techniques used.

The facial recognition process still has a lot to

be analyzed. Since several factors affect the

recognition of a face (age, lighting, gestures, scars,

distance, and angle to the camera, etc.).

Currently, we have programming languages

such as Python, which are incorporated into their

libraries some extraction and classification

techniques such as those studied in this work.

However, in the future, it would be interesting to

carry out facial recognition studies in 3D images. At

least the problem of lighting and angle to the camera

would be solved.

References:

[1] (text in Spanish) C. E. Franco, C. T. Ospina, E. S.

Cuevas, and D. V. Capacho, “Reconocimiento

Facial Basado En Eigenfaces, Lbhp Y Fisherfaces

En La Beagleboard-Xm,” Rev. Colomb. Tecnol.

Av., vol. 2, no. 26, 2017, doi:

10.24054/16927257.v26.n26.2015.2387.

[2] (text in Spanish) J. González Astudillo and M. G.

Zhindón Mora, “Plataforma de servicios de

reconocimiento facial para detección de prófugos

de la justicia en Ecuador,” Rev. Cienc. e Investig.,

vol. 5, pp. 31–41, 2020.

[3] (text in Spanish) M. Kuliah and M. Kuliah,

“Herramienta De Reconocimiento Facial Con

WSEAS TRANSACTIONS on SYSTEMS and CONTROL

DOI: 10.37394/23203.2023.18.5

José Cadena, Manuel Villa,

Maira Martínez, Jaime Acurio, Luis Chacón

E-ISSN: 2224-2856

Volume 18, 2023

Técnica De Visión Computacional 2d,” no. April,

pp. 33–35, 2019.

[4] F. Serratosa, “Biometria_ES_(Modulo_1),” p. 50,

2008.

[5] (text in Spanish) J. J. I. Casanova,

“Reconocimiento Facial En Ambientes No Autor :

Asesor : Línea de Investigación : Infraestructura ,

Tecnología y Medio Ambiente,” 2020.

[6] M. A. Turk and A. P. Pentland, “Face recognition

using eigenfaces,” Proceedings. 1991 IEEE

Comput. Soc. Conf. Comput. Vis. Pattern

Recognit., pp. 586–591, doi:

10.1109/CVPR.1991.139758.

[7] (text in Spanish) M. Centeno, “Tecnicas De

Biometria Basadas En Patrones Faciales Del Ser

Humano,” Экономика Региона, p. 32, 2012.

[8] A. Maisueche Cuadrado, “Utilización Del Machine

Learning En La Industria 4.0,” pp. 1–118, 2019,

[Online]. Available:

https://core.ac.uk/download/pdf/228074134.pdf#pa

ge=37&zoom=100,90,94.

[9] S. Shalev-Shwartz and S. Ben-David,

Understanding machine learning: From theory to

algorithms, vol. 9781107057. 2013.

[10] (text in Spanish) A. Moreno et al., Aprendizaje

automtico. 1994.

[11] D. Angluin, “Computational learning theory:

Survey and selected bibliography,” Proc. Annu.

ACM Symp. Theory Comput., vol. Part F1297, pp.

351–369, 1992, doi: 10.1145/129712.129746.

[12] (text in Spanish) P. F. De Máster, “Utilización de

métodos de visión artificial para PC como apoyo

en la automoción.,” 2015.

[13] “OpenCV: Clasificador en cascada.”

https://docs.OpenCV.org/3.4/db/d28/tutorial_casca

de_classifier.html (accessed Jan. 15, 2022).

[14] (text in Spanish) J. Ibarra-Estévez and K. Paredes,

“Redes neuronales artificiales para el control de

acceso basado en reconocimiento facial,”

Revistapuce, 2018, doi:

10.26807/revpuce.v0i106.140.

[15] S. Mullainathan and J. Spiess, “Machine learning:

An applied econometric approach,” J. Econ.

Perspect., vol. 31, no. 2, pp. 87–106, 2017, doi:

10.1257/jep.31.2.87.

[16] (text in Spanish) A. M. Llapapasca Montes and M.

A. Ochoa Zevallos, “Creación de una librería de

software de reconocimiento facial enfocado a la

identificación de trabajadores de una empresa

Ingeniería de Software,” Univ. Tecnológica del

Perú, 2019, [Online]. Available:

http://repositorio.utp.edu.pe/handle/UTP/2278.

[17] (text in Spanish) T. D. F. De Grado, “Vida En El

Reconocimiento Facial Jaime Varela de la

Escalera de Miguel,” 2019.

[18] R. Gottumukkal and V. K. Asari, “System level

design of real time face recognition architecture

based on composite PCA,” Proc. IEEE Gt. Lakes

Symp. VLSI, pp. 157–160, 2003, doi:

10.1145/764808.764849.

[19] (text in Spanish) J. D. Alvarado and J. Fernández,

“Análisis de textura en imágenes a escala de

grises, utilizando patrones locales binarios (LBP),”

2012.

[20] (text in Spanish) P. Álvarez, “Prototipo de sistema

piloto para control de acceso basado en

reconocimiento de rostros,” Fac. Ing. Univ. Mil.

Nueva Granada, p. 72, 2013.

[21] A. M. Martinez and A. C. Kak, “PCA versus

LDA,” IEEE Trans. Pattern Anal. Mach. Intell.,

vol. 23, no. 2, pp. 228–233, Feb. 2001, doi:

10.1109/34.908974.

[22] P. N. Belhumeur, J. P. Hespanha, and D. J.

Kriegman, “Eigenfaces vs. Fisherfaces:

Recognition using class specific linear projection,”

Lect. Notes Comput. Sci. (including Subser. Lect.

Notes Artif. Intell. Lect. Notes Bioinformatics), vol.

1064, no. 7, pp. 45–58, 1996, doi:

10.1007/bfb0015522.

[23] I. Challenger, Y. Díaz, and R. B. García, “El

lenguaje de programación Python/The

programming language Python,” Redalyc, vol. XX,

pp. 1–13, 2014, [Online]. Available:

https://www.redalyc.org/pdf/1815/181531232001.

pdf.

[24] (text in Spanish) La Biblioteca Estándar de Python

— documentación de Python - 3.10.2.

https://docs.python.org/es/3/library/index.html

(accessed Jan. 15, 2022).

Contribution of Individual Authors to the

Creation of a Scientific Article (Ghostwriting

Policy)

We confirm that all Authors equally contributed in

the present research, at all stages from the

formulation of the problem to the final findings and

solution.

Sources of Funding for Research Presented in a

Scientific Article or Scientific Article Itself

No funding was received for conducting this study.

Conflicts of Interest

Please declare anything related to this study.

Creative Commons Attribution License 4.0

(Attribution 4.0 International, CC BY 4.0)

This article is published under the terms of the

Creative Commons Attribution License 4.0

https://creativecommons.org/licenses/by/4.0/deed.en

_US

WSEAS TRANSACTIONS on SYSTEMS and CONTROL

DOI: 10.37394/23203.2023.18.5

José Cadena, Manuel Villa,

Maira Martínez, Jaime Acurio, Luis Chacón

E-ISSN: 2224-2856

Volume 18, 2023