Journal of Information System and Technology Research Volume 3, Issue 3, September 2024

Journal of Information System and Technology Research
journal homepage: https://journal.aira.or.id/index.php/jistr/

Direct implementation of AI-Based Facial Recognition for students
Andi Prayogi1*, Roy Francis Navea2, Rahmad Dian3, Muhammad Akbar Syabana Pane4, Ratu Mutiara Siregar5, Raden
Aris Sugianto6, Hasanal Fachri Satia Simbolon7
1,3,4,5,6,7

Department Information Systems and Technology, Institut Teknologi Sawit Indonesia, Indonesia
Department of Electronics and Computer Engineering, De La Salle University, Philippines

2

ARTICLE INFO

ABSTRACT

Article history:

The development of artificial intelligence (AI)-based facial recognition technology has become a
significant research topic in the field of computing and security. At the Indonesian Palm Oil Institute
(ITSI), AI-based facial recognition is introduced to students to improve their skills in developing AIbased applications. This study aims to implement and test a facial recognition system using a Python
program by utilizing a dataset generated independently. This research method involves several stages,
namely collecting ITSI students' facial data, data processing, creating a facial recognition model using
a machine learning algorithm, and evaluating model performance. The dataset used was developed
through a live shooting session involving active student participation. The facial recognition model
was trained using a convolutional neural network (CNN) algorithm that was optimized to improve
accuracy. The results of the study showed that the developed model was able to achieve high facial
recognition accuracy, with an average accuracy rate of 92%. The discussion includes an analysis of
factors that affect accuracy, such as variations in lighting and shooting angles, as well as the potential
use of this technology in a campus environment, including for attendance and security purposes. The
conclusion of this study shows that the implementation of AI-based facial recognition can be
effectively applied in an academic environment, as well as providing students with practical
experience in developing and testing AI applications. This study also opens up opportunities for further
research on improving the performance of facial recognition systems and their application in various
real-world scenarios.

Received August 13, 2024
Accepted September 30, 2024
Available online September 30,
2024
Keywords:
AI,
Face Recognition,
CNN
ITSI

© 2024 The Author(s). Published by AIRA.
This is an open access article under the CC BY-SA license
(http://creativecommons.org/licenses/by-sa/4.0/).

.

Corresponding Author:
Andi Prayogi,
Department Information Systems and Technology, Institut Teknologi Sawit Indonesia, Indonesia,
Street William Iskandar, Medan Estate, Medan City, 20226
Email: andiprayogi@itsi.ac.id

1.

INTRODUCTION

With the continuous advancement of science and technology, facial detection and recognition technology is
increasingly being applied in various fields, such as identity verification through facial scanning applications and unlocking
mobile phones. All these processes require facial detection and recognition technology. As technology becomes more diverse,
facial detection and recognition have become integral parts of our daily lives.
Research in the field of facial recognition highlights the importance of variations such as facial expressions, facial
shapes, and the distance between the eyes, which play a key role in distinguishing individual characteristics [1]. hen someone
looks at another person's face, they not only gather basic information such as age, gender, and ethnic background, but also form
subjective judgments regarding attractiveness, health, and personality. Aspects such as facial structure, smile, eyes, and lip
shape significantly influence how attractive someone is perceived to be by others [2].
One of the parts of humans that characterize recognition is the face. Full-face photos become an identity in class
attendance and sight on surveillance cameras [3]. The similarity of the shape of the face becomes a human characteristic as a
sign. In a similar facial shape, humans can consider them brothers even though they have different names and origins.Humans

107

JISTR, Volume 3, Issue 3, September 2024

P ISSN 2828-3864; E ISSN: 2828-2973

can distinguish someone who is born with twins and has a lot in common just from the shape of the face. The system of a
country identifying humans using the next face in the photo and then giving a number called NIK and name as label for data
collection [4].
Currently, the face has become a technology to open a security lock on a smartphone. The depiction of the human face
is a part that has its own complex form, so that the calculation system for the normal face shape model is a difficult calculation.
The weakness of face identification system is a problem in face recognition. Problem such as lighting, camera pixel size, hair
shape, use of glasses and masks, or various facial positions[5].
One characteristic that is often observed is the structural similarity between facial features and a person's emotional
expression. A face that appears calm but shows a slight resemblance to a happy expression (such as slightly upturned corners
of the mouth) is generally perceived as more trustworthy. Conversely, a neutral face with features resembling an angry
expression (such as lowered eyebrows) is often seen as dominant. Research also suggests that quick judgments about
trustworthiness or dominance often occur automatically and subconsciously. This underscores the importance of facial features
in shaping social perceptions, where a brief interpretation of someone's expression can influence social interactions and the
decisions we make about others. Therefore, understanding how facial features and emotional expressions interact can provide
valuable insights into facial recognition and human interaction. [6].
Computer vision technology is now widely applied in various systems, replacing humans in tasks involving pattern
recognition, such as in control, safety, forensics, medicine, and many other fields. This substitution occurs primarily because
computers offer a much higher level of efficiency. Unlike humans, machines do not require rest and are not affected by
emotional factors such as anger, fear, or boredom. With consistent reliability, computers can perform repetitive and complex
tasks without experiencing a decline in performance due to psychological or physical conditions[7]. One of the current
applications that use computer vision and image processing technology is an augmented reality based. Augmented reality turns
reality into a form of visualization in a supporting application such as 2D/3D animation, vector images, and its kind [8].This
modern application uses a facial recognition process like the social media applications Facebook and Instagram along with a
facial transfer application that can be used by AI in creating digital content. Furthermore, in the field of security, facial
recognition to unlock the door of the room has begun to be used. This situation was created to avoid many cases of theft that
are currently occurring in various places [9].
From the explanation above, further analysis will be carried out regarding the accuracy of the performance of the KNearest Neighbors Algorithm method in matching names based on facial characteristics for name recognition based on the
identity of the human face object.

2. RESEARCH METHOD
This research uses quantitative research, which is a type of systematic study of a phenomenon or situation by collecting
data that can be measured using statistical, mathematical, or computational techniques [10].
2.1. Training Data Used
The training data used in this study were students of the Information Systems and Technology Study Program.
Furthermore, each student took 50 photo frames from various positions and facial expressions so that the total training data used
was as many as the test objects. This activity is called Generate Dataset. The data from the photo capture is used as a dataset
for training data.
2.2 Data Processing
The facial recognition process consists of four main stages. The first stage is image preprocessing, where the
background of the image can be removed to focus on relevant areas. The second stage is post-processing, where the facial image
is resized to 128 x 128 pixels, with facial features properly aligned to ensure focus on areas such as the forehead, eyes, nose,
cheeks, mouth, and chin. The third stage involves feature extraction, using various image descriptors applied in existing
methods. Finally, the last stage is classification to determine identity[11].
The data obtained from the shooting session is then processed to improve image quality and reduce noise. This
processing includes face detection and cropping using Python libraries such as OpenCV and Dlib. After the faces are detected,
the images are normalized in size for consistency. Data augmentation is also performed to increase the number of datasets and
add variations, including rotation, cropping, and lighting adjustments.

108

JISTR, Volume 3, Issue 3, September 2024

P ISSN 2828-3864; E ISSN: 2828-2973

Figure 1. Research Flow In Data Processing
Figure 1 is a research flow for processing the received data. The following are the steps.
1. Start
This process marks the beginning of the research flow. At this stage, the system begins to initiate the next steps in
data processing.
2. Generated Dataset
At this stage, the system automatically generates a dataset consisting of 50 student photo images. This dataset will be
used as a basis for model training and testing. This process aims to collect image data that will be processed further.
Data is generated through an automated procedure that takes images from the initial data source.
3. Data Image
The first branch that is formed after the dataset is generated is Data Image. This process functions to manage the entire
dataset generated, namely 100% of the images from the previously generated dataset. All of this data will be used for
model training and evaluation.
4. Test Image
The second branch is Test Image, which is part of the generated dataset. At this stage, the dataset is divided, with 30%
of the images allocated for testing purposes. This separation ensures that the model will be tested with data that is
different from the training data, maintaining the objectivity of the test results.
5. Data Training
After the dataset separation, Data Training is the process in which the image data is trained using the selected
algorithm. At this stage, the dataset is processed to extract relevant features to build a classification model. The
learning algorithm will try to recognize important patterns in the image data to prepare for the classification stage.
6. Classification K-NN
At this stage, the training data from the previous process will be classified using the K-Nearest Neighbors (K-NN)
algorithm. This algorithm works by finding the nearest neighbor of each sample tested in the feature space, then
providing a classification label based on the majority of nearest neighbors. K-NN was chosen as the classification
algorithm because of its ability to handle distance and pattern-based classification.
7. Testing Accuracy
After the classification is complete, the system will proceed to the Testing Accuracy process, where the model will be
tested to measure its accuracy. This testing is done using the F1-Score metric, which combines precision and recall to
provide a better picture of accuracy, especially on imbalanced datasets. This process is important to evaluate the
performance of the model before being applied to wider data.

109

JISTR, Volume 3, Issue 3, September 2024

8.

P ISSN 2828-3864; E ISSN: 2828-2973

Finish
The last stage of this flow is Finish, which marks the end of the entire data processing process. Here, the test results
are reviewed and interpreted to determine the success of the model in classifying student images based on the data
generated.

2.3 Model Development
The facial recognition model was developed using the K-Nearest Neighbors (KNN) algorithm implemented in a
Python program with the Scikit-learn library. Feature extraction was performed using a histogram of oriented gradients
(HOG) to obtain a more easily analyzed facial representation. The KNN algorithm was then used to classify facial images
based on feature similarity. The model was trained using a dataset that had been processed with supervised learning
techniques.
K-Nearest Neighbors Classifier, is a method that has the function of classifying an object being tested where the
training data is stored. Here is the KNN formula using euclidean distance.
2
𝐷(𝑥, 𝑦) = √∑𝑁
𝑘=1(𝑋𝑘 − 𝑌𝑘 )

(1)

2.4 Test Evaluation
The evaluation of the built classification model is conducted through a series of precise measurements[12]. Model
performance is evaluated using four main parameters derived from the confusion matrix: metric precision, F1-score, precision,
and recall [13]. The confusion matrix for each model with different architectures is calculated at the point where the model
reaches optimal performance, providing a deeper understanding of the developed model's effectiveness[14]. Precision measures
how accurately the system's generated information is, ensuring that only correct and relevant information is classified
accurately[15]. Recall, on the other hand, assesses the system's ability to identify all correct information, providing insight into
the model's sensitivity. Accuracy reflects how close the identification results are to the actual values, serving as an important
indicator in evaluating the model's overall performance [16]. y using this combination of parameters, we can conduct a more
comprehensive analysis of the model's ability to perform classification accurately and consistently. The following is the formula
used for the evaluation stage.
Table 1. Table Evaluation
Prediction
Actual

Positive

Negative

Positive

TP

FN

Negative

FP

TN

Table 1 shows the results of the evaluation of the classification model using F1-Score, a metric that combines two main
components: Precision and Recall. F1-Score is an important metric because it provides a balance between precision and recall,
especially when the dataset used tends to be imbalanced. This table consists of four main elements, namely True Positive (TP),
True Negative (TN), False Positive (FP), and False Negative (FN), which are explained as follows:
1. True Positive (TP)
TP is the number of correct predictions where the model predicts the positive class (e.g., an image of a student that is
truly classified) and the prediction is correct. In the context of image recognition, this refers to cases where the model
successfully classifies an image correctly according to its original label. A high TP value indicates that the model can recognize
many samples correctly.
2. True Negative (TN)
TN is the number of correct predictions where the model predicts the negative class and the prediction is correct (e.g.,
an image of a student that is not included in a certain classification and the prediction is correct). This indicates that the model
is able to correctly distinguish images that are not included in the target class. A high TN value indicates that the model is good
at excluding incorrectly classified images.

110

JISTR, Volume 3, Issue 3, September 2024

P ISSN 2828-3864; E ISSN: 2828-2973

3. False Positive (FP)
FP, or commonly called "false positive", is the number of cases where the model incorrectly predicts the positive class,
when in fact the sample does not belong to that class. In this case, the image is classified as positive by the model, but in reality
it is not. A high FP value indicates that the model makes many errors when classifying images that should not be in a particular
class.
4. False Negative (FN)
FN, or "false negative", occurs when the model fails to predict the positive class, and instead classifies an image that
is actually positive as negative. This means that the model is unable to detect samples that should be included in the target
classification. A high FN value indicates that the model has difficulty in accurately identifying positive samples.
From this table, the F1-Score can be calculated as a combined measure that takes into account both Precision and
Recall. Precision is calculated from the ratio of TP to the sum of TP + FP (how many positive predictions were correct), while
recall is calculated from the ratio of TP to the sum of TP + FN (how many positive samples the model successfully recognized).
The F1-Score then provides the result in the form of a single value, which is very useful for evaluating model performance,
especially if there is imbalance in the data.
The following are explanations of some of the terms mentioned above. TP (True Positive) refers to the number of cases
where the actual class is positive, and the prediction result is also positive. FN (False Negative) refers to the number of cases
where the actual class is positive, but the prediction is negative. FP (False Positive) indicates the number of cases where the
prediction shows positive, even though the actual class is negative. TN (True Negative) is the number of cases where the actual
class is negative, and the prediction is also negative. [14].
𝑇𝑃 + 𝑇𝑁

𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 = 𝑇𝑃+ 𝑇𝑁 +𝐹𝑁+𝐹𝑃

(2)

3. RESULTS AND DISCUSSION
3.1 Results
This program is written using the Python programming language. After the webcam is run, it will automatically open
the application in windows as an indicator that the program has been running and has successfully carried out the face
detection process for the capture process. After the capture process is collected, it will be tested first using images as the
testing phase. Furthermore, if the testing phase is complete, it will be tested using a webcam camera directly. The program
results are shown in Figure 2.

Figure 2. Program Result View
This study produces a facial recognition system based on the K-Nearest Neighbors (KNN) algorithm with an average
accuracy of 92% on test data. Testing was carried out under several conditions: images of individual faces facing forward,

111

JISTR, Volume 3, Issue 3, September 2024

P ISSN 2828-3864; E ISSN: 2828-2973

images of two people with normal lighting, and images of more than two people in dim lighting conditions. Out of a total of
500 test images, the model successfully classified 460 images correctly.
The results show that the KNN model is able to recognize faces well under various conditions, although there are some
challenges related to lighting and the number of subjects in the image.

3.2 Process Analysis
The simple logic of KNN is to explore its immediate surroundings, assume the same test points as them and get the
results. The advantage of this KNN method is that the machine learning model is easy and simple among other machine learning
methods, so it is very suitable for classifying models that have predetermined values through the training process. Face
recognition program (face detection) after training the program, the program will produce information whether the face is known
or unknown with an accuracy value that matches the results of face detection. The testing process involves recognizing faces
from images taken via webcam. The program is tested under several conditions: images of individual faces facing forward,
images of two people with normal lighting, and images of more than two people in dim lighting conditions. The front view
condition can be seen in Figure 3

Figure 3. Front View of a Student
3.3 Experimenting with Many Conditions
Testing was carried out under several conditions to assess the robustness and flexibility of the model in different
situations:
1. Front-Facing Individual Face Image: In this condition, the model achieved 95% accuracy, indicating that facial feature
recognition is highly accurate when there is only one subject with their face facing directly towards the camera. This condition
is the most ideal because all facial features can be clearly detected without interference from other subjects or lighting variations.
The condition can be seen in figure 4.

Figure 4. Front-Facing Individual Face Image

112

JISTR, Volume 3, Issue 3, September 2024

2.

P ISSN 2828-3864; E ISSN: 2828-2973

Two-Person Image with Normal Lighting: The model recorded 90% accuracy in this condition. Normal lighting allows
the model to detect and recognize facial features well even when there is more than one subject. However, having more
than one face in an image starts to present challenges in distinguishing similar features between subjects. The condition can
be seen in figure 5.

Figure 5. Two-Person Image
3.

Multiple-Person Images in Low-Light Conditions: Accuracy drops to 85% in this condition, indicating that the model
has more difficulty detecting and recognizing faces when the lighting is poor and there are multiple subjects. Low lighting
reduces the contrast and clarity of facial features, while the presence of multiple subjects introduces more noise and
complexity into the classification. The condition can be seen in figure 6, Figure 7, Figure 8 and Figure 9.

Figure 6. Multiple Person Images with Low-Light
In Figure 6, the system faces the challenge of recognizing the faces of multiple individuals in low-light conditions.
This condition can reduce the image quality because facial details become less clear, making it difficult for the facial recognition
algorithm to extract accurate features. This affects the model's performance in correctly identifying faces. Low lighting also
often introduces noise into the image, which can potentially decrease classification accuracy.

113

JISTR, Volume 3, Issue 3, September 2024

P ISSN 2828-3864; E ISSN: 2828-2973

Figure 7. Multiple Person Images with High-Light
Figure 7 shows several individuals in high-light conditions, where facial details are more clearly visible compared to
low-light conditions. In good lighting, facial features such as contours, skin textures, and face shapes are more easily extracted
by the facial recognition algorithm. However, even in bright lighting, if the lighting angle is too sharp or overexposure occurs,
important details can still be lost, which can also affect classification accuracy.

Figure 8. Multiple Person images and Not Correct
In Figure 8, the classification results show errors in recognizing the names of several individuals. This indicates that
the facial recognition model has difficulty in accurately identifying some individuals. Such errors can be caused by various
factors such as similarity in facial features between individuals, variations in poses, or facial expressions that do not match the
training data. These prediction errors are important to note as part of the overall evaluation of the model's performance.

114

JISTR, Volume 3, Issue 3, September 2024

P ISSN 2828-3864; E ISSN: 2828-2973

Figure 9. Bright Lighting Side View Face Image
Figure 9 shows an individual’s face taken from a side angle with bright lighting. The side view of a face presents a
challenge for facial recognition algorithms, as many important features such as both eyes, nose shape, and facial symmetry are
not fully visible. While bright lighting can enhance details, this angle reduces the amount of information the model can extract,
potentially lowering recognition accuracy.This test illustrates how environmental conditions and subject configuration can
significantly affect facial recognition performance.

3.4 Factors Affecting Accuracy
Various factors have been identified that influence model accuracy under each condition tested, and understanding
these factors is important for further improvement:
1. Lighting: Good lighting is essential to ensure facial features are clearly detected. Poor lighting can cause shadows
that obscure important details of facial features, while too bright lighting can cause overexposure, which also hides
details.
2. Number of Subjects: The more subjects in an image, the more complex the facial recognition task becomes. The
model must be able to detect and isolate each face before performing recognition, and this can lead to confusion,
especially if the faces have similar features or are adjacent to each other.
3. Image Quality: Low image quality, such as poor resolution or noise, reduces the model’s ability to detect features
accurately. Clear, high-quality images greatly improve accuracy because more information can be extracted by the
algorithm.
4. Facial Expression: Variations in facial expression can change the geometric and textural features detected by the
model, which can affect recognition accuracy. Facial expressions that are very different from the training images can
lead to misclassification.
5.

3.5 Discussion
The results show that the KNN algorithm is effective for face recognition under various conditions, especially in
academic environments. The advantages of this model in terms of simplicity and processing speed make it suitable for
applications such as face-based attendance and access control systems. However, for face recognition in poor lighting conditions
or with many subjects, further improvements are needed. Further development can include testing with larger datasets and
integrating other machine learning techniques to improve the robustness and accuracy of the system. Exploration of more
sophisticated image processing methods can also improve reliability under challenging test conditions.

4. CONCLUSION
In this study, we successfully developed a facial recognition system based on the K-Nearest Neighbors (KNN)
algorithm that achieved an average accuracy of 92% across a range of test conditions. The model performed well in recognizing
individual faces, especially when the subject was facing directly towards the camera. A greater challenge arose when lighting
conditions were reduced and there were multiple subjects in the image. Through comprehensive data collection and careful data

115

JISTR, Volume 3, Issue 3, September 2024

P ISSN 2828-3864; E ISSN: 2828-2973

processing, the system was able to accurately recognize facial features, and the KNN algorithm proved to be efficient and fast
for the task of face classification. However, accuracy was still affected by image quality, lighting variations, and the number of
subjects, indicating that there is room for further improvement. As facial recognition technology advances today, the results of
this study highlight the potential for using AI-based systems in academic environments such as campuses to improve efficiency
and security in attendance and access management. The speed and simplicity of KNN make it a suitable choice for small to
medium-scale applications. However, for more complex and large-scale applications, the use of more sophisticated machine
learning techniques such as deep learning can provide further improvements in the accuracy and robustness of the system.

REFERENCES
H. Ge, Z. Zhu, R. Liu, and X. Wu, “Application of Multiscale Facial Feature Manifold Learning Based on VGG-16,” J.
Sensors, vol. 2021, 2021, doi: 10.1155/2021/7129800.
[2] G. Kanavakis, D. Halazonetis, C. Katsaros, and N. Gkantidis, “Facial shape affects self-perceived facial attractiveness,”
PLoS One, vol. 16, no. 2 February, pp. 1–17, 2021, doi: 10.1371/journal.pone.0245557.
[3] M. A. S. Pane, K. Saleh, A. Prayogi, R. Dian, R. M. Siregar, and R. Aris Sugianto, “Low-Cost CCTV for Home Security
With Face Detection Base on IoT,” J. Inf. Syst. Technol. Res., vol. 3, no. 1, pp. 20–29, 2024, doi: 10.55537/jistr.v3i1.769.
[4] G. A. Rama, F. Fauziah, and N. Nurhayati, “Perancangan Sistem Keamanan Brangkas Menggunakan Pengenalan Wajah
Berbasis Android,” J. Media Inform. Budidarma, vol. 4, no. 3, p. 635, 2020, doi: 10.30865/mib.v4i3.2149.
[5] A. Prayogi et al., “ENHANCING NETWORK PERFORMANCE LOAD BALANCING IN CYBER CAFE NETWORKS
WITH DIJKSTRA ALGORITHM ON MIKROTIK,” J. Tek. Inform., vol. 5, no. 1, pp. 253–261, 2024, [Online].
Available: https://jutif.if.unsoed.ac.id/index.php/jurnal/article/view/1644/448
[6] B. Jaeger and A. L. Jones, “Which Facial Features Are Central in Impression Formation?,” Soc. Psychol. Personal. Sci.,
2021, doi: 10.1177/19485506211034979.
[7] R. Grycuk, A. Wojciechowski, W. Wei, and A. Siwocha, “Detecting Visual Objects by Edge Crawling,” J. Artif. Intell.
Soft Comput. Res., vol. 10, no. 3, pp. 223–237, 2020, doi: 10.2478/jaiscr-2020-0015.
[8] F. Gonten, F. Nfwan, and A. Ya’u Gital, “Pre-Review Convolutional Neural Network for Detecting Object in Image
Comprehensive Survey and Analysis,” J. Inf. Syst. Technol. Res., vol. 3, no. 2, pp. 45–64, 2024, doi:
10.55537/jistr.v3i2.799.
[9] A.- Suryansah, R.- Habibi, R. M. Awangga, and R. N. S. Fatonah, “Implementasi Face Recognition Untuk Mengakses
Ruangan,” J. Mediat., vol. 3, no. 3, p. 25, 2020, doi: 10.26858/jmtik.v3i3.15176.
[10] “Jenis-Jenis Penelitian Lengkap dengan Contoh dan Penjelasannya.” Accessed: Feb. 14, 2022. [Online]. Available:
https://penerbitdeepublish.com/jenis-jenis-penelitian/
[11] M. Z. Syahputra and M. D. Irawan, “Application Of Bootstrap In Building Application Of Fertilizer Inventory System,”
J.
Inf.
Syst.
Technol.
Res.,
vol.
1,
no.
1,
2022,
[Online].
Available:
http://download.garuda.kemdikbud.go.id/article.php?article=2819386&val=25211&title=Application of Bootstrap in
Building Fertilizer Inventory System Applications
[12] L. Farokhah, “Implementasi K-Nearest Neighbor untuk Klasifikasi Bunga Dengan Ekstraksi Fitur Warna RGB,” J. Teknol.
Inf. dan Ilmu Komput., vol. 7, no. 6, p. 1129, 2020, doi: 10.25126/jtiik.2020722608.
[13] Muhammad Nur Ichsan, Nur Armita, Agus Eko Minarno, Fauzi Dwi Setiawan Sumadi, and Hariyady, “Increased
Accuracy on Image Classification of Game Rock Paper Scissors using CNN,” J. RESTI (Rekayasa Sist. dan Teknol.
Informasi), vol. 6, no. 4, pp. 606–611, Aug. 2022, doi: 10.29207/resti.v6i4.4222.
[14] M. Harahap, Em Manuel Laia, Lilis Suryani Sitanggang, Melda Sinaga, Daniel Franci Sihombing, and Amir Mahmud
Husein, “Deteksi Penyakit Covid-19 Pada Citra X-Ray Dengan Pendekatan Convolutional Neural Network (CNN),” J.
RESTI (Rekayasa Sist. dan Teknol. Informasi), vol. 6, no. 1, pp. 70–77, 2022, doi: 10.29207/resti.v6i1.3373.
[15] R. I. Borman, Y. Fernando, and Y. Egi Pratama Yudoutomo, “Identification of Vehicle Types Using Learning Vector
Quantization Algorithm with Morphological Features,” J. RESTI (Rekayasa Sist. dan Teknol. Informasi), vol. 6, no. 2, pp.
339–345, Apr. 2022, doi: 10.29207/resti.v6i2.3954.
[16] I. Weni, P. E. P. Utomo, B. F. Hutabarat, and M. Alfalah, “Detection of Cataract Based on Image Features Using
Convolutional Neural Networks,” IJCCS (Indonesian J. Comput. Cybern. Syst., vol. 15, no. 1, p. 75, 2021, doi:
10.22146/ijccs.61882.
[1]

116