Restoration and Text Extraction for Enhanced Analysis of Vintage
and Damaged Documents Using Deep Learning
Anand Magar, Rutuja Desai, Siddhi Deshmukh, Samarth Deshpande and Sakshi Dhamne
Department of Computer Engineering, Vishwakarma Institute of Technology, Pune, India
Keywords: Deep Learning, DE-GAN, OCR, Image Restoration, Text Extraction.
Abstract: Restoration and analysis of vintage and damaged documents are crucial in preserving valuable historical and
cultural records. In this paper, we present a novel approach that combines Document Enhancement Generative
Adversarial Network (DE-GAN) for image restoration with Optical Character Recognition (OCR) for
efficient text extraction. Our methodology focuses on restoring degraded documents by enhancing visual
quality and legibility, allowing for more accurate text retrieval. By employing DE-GAN, we can mitigate
various forms of document degradation, such as stains, faded ink and physical damage, while the OCR system
effectively extracts the underlying text for further analysis. The proposed framework enhances the quality of
historical document preservation and facilitates better data retrieval, ensuring that critical information is not
lost due to document deterioration. Experimental results demonstrate significant improvements in both
restoration quality and text extraction accuracy, making our method a robust solution for historical document
analysis.
1 INTRODUCTION
This Historical and archival documents are an
invaluable part of our collective heritage, but over
time, many of these documents suffer from
deterioration caused by environmental factors,
physical damage, or improper storage. Such
degradation often results in the loss of crucial
information and poses a significant challenge to
researchers and archivists who aim to preserve these
documents for future generations. Traditional
restoration techniques, while effective, can be labor-
intensive and time-consuming. Moreover, manual
text extraction from these degraded documents is
prone to errors and inconsistencies.
Recent advances in deeplearning and computer
vision have opened up new possibilities for
automating the restoration and text extraction
process. Generative Adversarial Networks (GANs),
in particular, have shown great promise in various
image restoration tasks due to their ability to generate
high-quality images from degraded inputs. In this
study, we introduce a DE-GAN-based approach to
restore damaged documents, enhancing both their
visual clarity and structural integrity. Paired with
OCR technology, this enables the accurate extraction
of text from the restored images, even in cases where
traditional OCR techniques would struggle due to
document quality.
The combination of DE-GAN for restoration and
OCR for text extraction offers a comprehensive
solution to the challenges posed by document
degradation. Our approach not only helps preserve
the visual quality of historical documents but also
facilitates easier access to the text they contain,
enabling researchers and archivists to perform more
in-depth analysis on restored documents.
2 LITERATURE SURVEY
Literature survey has been carried out to understand
innovative approaches.
Researchers have used deep learning method
image inpainting and then valued performance of the
model to identify how an algorithm can restore an
image using many performance metrics. utilization of
available power. EMS is crucial for optimal power
balance in hybrid PV/Wind turbine systems. The
objective of EMS is for improving transients, MPPT,
EMS for grid. The flow chart for an EMS is shown in
708
Magar, A., Desai, R., Deshmukh, S., Deshpande, S. and Dhamne, S.
Restoration and Text Extraction for Enhanced Analysis of Vintage and Damaged Documents Using Deep Learning.
DOI: 10.5220/0013600500004664
Paper published under CC license (CC BY-NC-ND 4.0)
In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 2, pages 708-714
ISBN: 978-989-758-763-4
Proceedings Copyright © 2025 by SCITEPRESS – Science and Technology Publications, Lda.

Figure 4.. Authors have done a survey of the
applications of deep learning in the field of image
restoration and gives some ways for future research.
(Shubekova, Beibitkyzy, et al. , 2023). Authors have
used deep convolutional neural networks for
research. Research include four restoration tasks
include image inpainting, pixel interpolation, image
deblurring and image denoising and used three
different datasets. Results demonstrate that approach
performs well on all models. (Gao and Grauman,
2017). In another study, two neural network
structures were used to estimate background light and
scene depth to enhance underwater image restoration.
Experimental results on both synthetic and real
underwater images highlight the effectiveness of the
method. (Cao, Peng, et al. , 2018). Researchers
developed a convolutional neural network (CNN)
model to de-haze individual images, facilitating
further restoration and improvement. The model's
performance was evaluated using images and features
from distinct regions, demonstrating its
generalization capabilities. (Ramkumar, Ayyadurai,
et al. , 2021). The development of speckle reduction
techniques has advanced alongside progress in image
restoration, with deep neural networks offering new
solutions. These networks have surpassed earlier
methods such as patches, sparsity, wavelet
transforms, and total variation minimization in terms
of restoration performance. (Denis, Dalsasso, et al. ,
2021). A proposed method for digital mural
restoration blocks high-priority sample blocks,
prevents processing of areas with numerous unknown
pixels, and minimizes error accumulation, achieving
over 20% accuracy improvement compared to
baseline methods. This technique excels in restoring
the main structure and texture details of complex
images with significant information loss. (Xiao,
Zheng, et al. , 2023). A generative adversarial
network (GAN) with embedded channels and spatial
attention was introduced, increasing the perceptual
field through expanded convolution. This allows the
network to better capture image details and broken
edges, leading to enhanced restoration performance,
as shown by experimental results on public datasets.
(Meng, Yu, et al. , 2022). A comparative study of
methods for restoring noisy images, including the
Wiener filter, wavelet methods, and Wiener filtering
with BM3D techniques, was also conducted. The
study analyzed the effectiveness of different color
models like RGB and YCbCr in image segmentation,
alongside CNN-based image classification,
considering factors such as activation functions and
pooling methods. (Vispute, Rajeswari, et al. , 2023).
Lastly, a CNN regression model was employed to
learn enhancement parameters for different types of
underwater images. Quantitative and qualitative
experiments conducted on known datasets
demonstrated impressive accuracy rates, achieving
better results on PSNR and SSIM quality metrics
compared to other techniques. (Martinho,
Calvalcanti, et al. , 2024). Researchers in the field of
handwritten text recognition (HTR) have focused on
improving accuracy using deep learning models like
LSTM. In one study, an LSTM-based HTR model,
integrated with an OCR system, outperformed
alternative methods on the IAM handwritten dataset,
with the 2DLSTM approach showing superior results.
(Nikitha, Geetha, et al. , 2020). A study on text
recognition from real-time images compared two
feature extraction methods: Histogram of Oriented
Gradients (HOG) and Local Binary Patterns (LBP).
After preprocessing and text localization, a CNN-
based model was used for text recognition. LBP
proved more effective in accuracy, precision, and F1
score compared to HOG. (Harsha, Kumar, et al. ,
2022). Text recognition from natural images remains
challenging due to complex backgrounds and
irregular text arrangements. A deep learning model
(DL-TRI) was developed to handle curved and
perspective fonts, addressing these complexities more
effectively than previous methods. (Shrivastava,
Amudha, et al. , 2019). Handwriting recognition
systems, which previously relied on optical character
recognition (OCR), now utilize CNNs, LSTMs, and
Connectionist Temporal Classification (CTC) for
better accuracy. This approach was tested on the
MNIST dataset using OpenCV and TensorFlow for
image processing and word recognition, yielding
improved performance. (Ansari, Kaur, et al. , 2022).
Researchers are recognizing handwritten text, a two-
phase method using CNNs was proposed. The first
phase identifies the input, while the second phase
classifies the language. The system, tested on the
MNIST database, achieved 99% testing accuracy and
99.6% training accuracy. (Thilagavathy, Suresh, et al.
, 2023). The architecture, design, and testing of a
Handwritten Character Recognition System are
presented, demonstrating the efficiency of neural
networks in recognizing handwritten characters. The
system converts images of handwritten notes from
students and instructors into digital text for further
use. (MS and A. D. R, 2023). Deep learning and
machine learning techniques have shown significant
promise in optical character recognition (OCR). A
comprehensive overview of four key architectures—
Support Vector Machines (SVM), Artificial Neural
Networks (ANN), Naive Bayes, and Convolutional
Neural Networks (CNN)—is provided, highlighting
Restoration and Text Extraction for Enhanced Analysis of Vintage and Damaged Documents Using Deep Learning
709

their roles in improving OCR. (Sharma, Kaushik, et
al. , 2020). The evolution of OCR due to
advancements in deep learning algorithms is
discussed, focusing on the introduction of
Convolutional Recurrent Neural Networks (CRNN)
combined with attention mechanisms to enhance text
recognition performance. (N. J and S. K. A M ,
2024). The increasing availability of GPUs and cloud
platforms such as Google Cloud and Amazon Web
Services has expanded computational power,
facilitating the training of complex neural networks.
This study details the design of an image
segmentation-based handwritten character
recognition system using OpenCV for image
processing and TensorFlow for neural network
training, implemented in Python. (Vaidya, Trivedi, et
al. , 2018). Handwritten Text Recognition (HTR) has
garnered significant attention due to its broad
applications, yet it remains a challenging research
area. The variability in personal handwriting and the
characteristics of handwritten characters across
different languages pose difficulties for HTR
systems. Traditional approaches rely on
Convolutional Neural Networks (CNNs) and
Recurrent Neural Networks (RNNs) with the
Connectionist Temporal Classification (CTC)
objective function. However, attention-based
sequence-to-sequence (Seq2Seq) models have
emerged as a more flexible solution, particularly for
the temporal nature of text. These models utilize
attention mechanisms to focus on the most relevant
features of the input, offering enhanced adaptability.
This paper provides a comprehensive comparison of
deep learning approaches for HTR and discusses the
challenges that limit their effectiveness. (Teslya,
Mohammed, et al. , 2022).
3 METHODOLOGY
The methodology of this research paper focuses on
the use of Document Enhancement Generative
Adversarial Network (DE-GAN) for restoring
vintage and damaged documents, coupled with
Optical Character Recognition (OCR) to extract text
for analysis. The methodology is divided into the
following steps: dataset preparation, GAN
architecture, training process, and evaluation.
3.1 Dataset Preparation
We use a set of degraded document images from the
DIBCO datasets (2009-2017) for training and
evaluation. The dataset consists of two types of
images.
• Degraded images (A): These are the input
images affected by various forms of damage
such as stains, fading, and noise.
• Clean images (B): These images serve as the
ground truth for the restoration process.
Before feeding the images into the model, the
degraded and clean images are resized to a uniform
size of 256 × 256 × 1 to ensure compatibility with the
GAN network architecture. Preprocessing is applied
to convert the images to grayscale to simplify the
restoration task.
3.2 GAN Architecture
The restoration process is carried out using a
Generative Adversarial Network (GAN), which
consists of a generator and a discriminator:
• Generator Model: The generator's task is to
restore degraded document images to their
original, clean state. It takes as input a
degraded image and outputs a restored
version. The generator is built using a
convolutional neural network with multiple
layers, gradually increasing the resolution of
the generated images. The largest layer in the
generator has 1024 units to handle complex
restorations.
• Discriminator Model: The discriminator's
job is to distinguish between real clean
images and those generated by the generator.
It learns to classify the restored images as
either real (clean) or fake (restored). This
binary classification helps the generator
improve over time. The discriminator is also
built using a convolutional architecture and
is periodically trained on both the real and
generated images.
The generator and discriminator models are
combined into a GAN network, where the generator
tries to fool the discriminator by generating realistic
images, and the discriminator attempts to classify
them correctly.
3.3 Training Process
The GAN training is performed as follows:
Adversarial Training: The generator and
discriminator are trained in alternating steps. In each
epoch, the following procedure is followed:
INCOFT 2025 - International Conference on Futuristic Technology
710

• Load Data: For each image in the
dataset, both degraded and clean images
are loaded from respective directories
(e.g., data/A/ for degraded images and
data/B/ for clean images).
• Image Conversion and Preprocessing:
The images are converted to grayscale,
resized, and split into smaller patches of
256 × 256. This helps ensure that the
network can focus on smaller details
within the document.
• Batch Creation: Mini-batches of size
128 are created for training the generator
and discriminator. Each mini-batch
contains pairs of degraded and clean
patches.
• Training the Discriminator: The
discriminator is trained on two batches
of images: the clean images (real) and
the generator’s output (fake). The goal is
to correctly classify the real and fake
images. The loss function for the
discriminator is binary cross-entropy.
• Training the Generator: The generator
is trained to generate images that can
deceive the discriminator into
classifying them as real. The GAN loss
function combines pixel-wise
reconstruction loss (to ensure accuracy
in restoration) and adversarial loss (to
ensure the images appear realistic).
• Model Weight Saving: The generator
and discriminator weights are saved
periodically to ensure continuity in
training and for future use.
The training is conducted over 80 epochs, with the
generator and discriminator weights saved at each
epoch for potential further use. The model is trained
using the Adam optimizer with a small learning rate
to stabilize the GAN’s learning process.
3.4 Prediction and Inference
After training, the generator is used to predict
restored images from unseen degraded images:
3.4.1 Input Image Preprocessing:
The degraded images are padded and resized to
ensure they are divisible into 256 × 256 patches. The
patches are fed into the trained generator for
restoration.
3.4.2 Image Stitching:
The restored patches are merged back into the full
document, reconstructing the entire image. The final
restored image is saved as output.
3.4.3 Text Extraction:
After restoration, OCR tools are applied to extract
text from the restored images, enabling document
analysis and data retrieval.
3.5 Evaluation
To evaluate the restoration quality, we use the
following metrics:
• Peak Signal-to-Noise Ratio (PSNR): This
measure the accuracy of the restored image
compared to the ground truth. A higher
PSNR indicates better restoration quality.
• Structural Similarity Index (SSIM): SSIM
evaluates the perceptual quality of the
restoration by comparing structural features
between the restored and clean images.
Additionally, for each epoch, we evaluate the
generator’s output using OCR-based text extraction
to measure the improvement in legibility and
accuracy of text recognition.
The methodology thus combines the strengths of
GAN-based restoration with text extraction using
OCR, providing a powerful framework for enhancing
and analyzing damaged document images.
4 RESULTS AND DISCUSSION
The performance of the proposed DE-GAN model for
document restoration was evaluated on the DIBCO
datasets (2009-2017). The model was trained over 80
epochs, and the restored document images were
compared with their ground truth counterparts using
two key metrics: Peak Signal-to-Noise Ratio
(PSNR) and Structural Similarity Index (SSIM).
4.1 Quantitative Evaluation
The PSNR and SSIM values were calculated for each
restored image and averaged across the test set. The
results demonstrated significant improvement in the
restoration quality of the degraded document images:
• PSNR: The average PSNR across the
validation set was 30.25 dB, indicating a
high level of fidelity between the restored
Restoration and Text Extraction for Enhanced Analysis of Vintage and Damaged Documents Using Deep Learning
711

images and the clean ground truth images.
This suggests that DE-GAN successfully
removed noise, stains, and other
degradation artifacts while preserving
essential document details.
• SSIM: The SSIM score averaged 0.92,
reflecting that the restored images retained
much of the structural content of the
original clean documents. The model
effectively handled the recovery of fine
details, such as text edges and document
structure, even in highly degraded areas.
These results confirm that DE-GAN
significantly outperforms traditional document
restoration methods in terms of both perceptual
quality and pixel-level accuracy.
4.2 Qualitative Evaluation
Qualitatively, the restored images exhibited notable
improvements in terms of visual clarity and legibility.
Before restoration, many document images were
heavily degraded by noise, fading, and ink bleeding,
making it difficult to extract readable text. After
restoration with DE-GAN:
• Text clarity improved substantially,
allowing for easier visual reading of both
printed and handwritten text.
• Noise and stains were effectively removed
from the images without overly smoothing
or distorting the text.
• Fine details, such as small font sizes and
intricate document structures, were well-
preserved in the restoration process.
Several sample outputs showed that even in
severely degraded regions, DE-GAN was able to
generate visually appealing and structurally
accurate document images.
4.3 OCR Text Extraction Performance
The integration of OCR for text extraction was
performed on the restored document images, and the
results were compared with the OCR performance on
the original degraded images.
4.3.1 OCR Accuracy
The Character Error Rate (CER) and Word Error
Rate (WER) were used to measure the accuracy of
text extraction:
• CER: The average CER after restoration
was reduced by 40% compared to the
original degraded images, with the final
CER being 6.5%. This indicates a
substantial reduction in the number of
misrecognized characters, thanks to the
improved legibility of the restored images.
• WER: The average WER after restoration
dropped to 8.2%, representing a 35%
improvement over the baseline performance
on degraded images. This reflects more
accurate word-level recognition, especially
in documents with heavily damaged text.
The improvement in both CER and WER
highlights the effectiveness of DE-GAN in
enhancing the quality of document restoration,
enabling more accurate and reliable text extraction
using OCR.
4.3.2 Impact on Historical Document
Analysis
In the context of vintage and historical documents,
where even partial degradation can hinder text
recognition, the DE-GAN model proved invaluable.
The restored documents not only facilitated better
OCR performance but also allowed for manual
inspection of legible text, preserving valuable
historical content that would otherwise be lost or
inaccessible.
Figure 1: Original Image
INCOFT 2025 - International Conference on Futuristic Technology
712

Figure 2: Restored Image
Figure 3: Original Handwritten Image
Figure 4: Restored Handwritten Image
5 CONCLUSIONS
This research introduced a highly effective DE-GAN
framework for restoring degraded and damaged
documents, resulting in significant improvements in
both image quality and text extraction accuracy. By
training the model on the DIBCO 2009-2017 datasets,
DE-GAN demonstrated superior performance in
recovering clear and structurally accurate document
images, as reflected in the high average PSNR of
30.25 dB and SSIM of 0.92. These metrics
underscore the framework's ability to restore visual
clarity and preserve fine textual details, leading to
enhanced document legibility.
Moreover, the integration of Optical Character
Recognition (OCR) within the DE-GAN framework
yielded a substantial reduction in text recognition
errors. The Character Error Rate (CER) decreased by
40%, bringing the average CER to 6.5%, while the
Word Error Rate (WER) improved by 35%, dropping
to 8.2%. These improvements demonstrate the
model's capability to significantly enhance OCR
accuracy, especially in documents with heavy
degradation or complex text structures.
While DE-GAN outperformed traditional
restoration techniques across various levels of
degradation, challenges remain in handling severely
damaged regions and complex document layouts.
Future work will aim to further enhance the model's
generalization to diverse document types, develop
automated text correction pipelines, and improve
real-time processing.
REFERENCES
R. K. Cho, K. Sood and C. S. C. Channapragada, "Image
Repair and Restoration Using Deep Learning," 2022
4th International Conference on Artificial Intelligence
and Speech Technology (AIST), Delhi, India, 2022, pp.
1-8, doi: 10.1109/AIST55798.2022.10065203.
A. Shubekova, A. Beibitkyzy and A. Makhazhanova,
"Application of Deep Learning in the Problem of Image
Restoration," 2023 IEEE International Conference on
Smart Information Systems and Technologies (SIST),
Astana, Kazakhstan, 2023, pp. 158-163, doi:
10.1109/SIST58284.2023.10223540.
R. Gao and K. Grauman, "On-demand Learning for Deep
Image Restoration," 2017 IEEE International
Conference on Computer Vision (ICCV), Venice, Italy,
2017, pp. 1095-1104, doi: 10.1109/ICCV.2017.124.
K. Cao, Y. -T. Peng and P. C. Cosman, "Underwater Image
Restoration using Deep Networks to Estimate
Background Light and Scene Depth," 2018 IEEE
Southwest Symposium on Image Analysis and
Restoration and Text Extraction for Enhanced Analysis of Vintage and Damaged Documents Using Deep Learning
713

Interpretation (SSIAI), Las Vegas, NV, USA, 2018, pp.
1-4, doi: 10.1109/SSIAI.2018.8470347.
G. Ramkumar, A. G, S. K. M, M. Ayyadurai and S. C, "An
Effectual Underwater Image Enhancement using Deep
Learning Algorithm," 2021 5th International
Conference on Intelligent Computing and Control
Systems (ICICCS), Madurai, India, 2021, pp. 1507-
1511, doi: 10.1109/ICICCS51141.2021.9432116.
L. Denis, E. Dalsasso and F. Tupin, "A Review of Deep-
Learning Techniques for SAR Image Restoration,"
2021 IEEE International Geoscience and Remote
Sensing Symposium IGARSS, Brussels, Belgium,
2021, pp. 411-414, doi:
10.1109/IGARSS47720.2021.9555039.
H. Xiao, H. Zheng and Q. Meng, "Research on Deep
Learning-Driven High-Resolution Image Restoration
for Murals From the Perspective of Vision Sensing," in
IEEE Access, vol. 11, pp. 71472-71483, 2023, doi:
10.1109/ACCESS.2023.3295253.
Y. Li, J. Meng, Y. Yu, C. Wang and Z. Guan, "Image
Restoration Based on Improved Generative Adversarial
Networks," 2022 7th International Conference on
Image, Vision and Computing (ICIVC), Xi’an, China,
2022, pp. 799-804, doi:
10.1109/ICIVC55077.2022.9886285.
S. R. Vispute, K. Rajeswari, A. Nema, A. Jagtap, M.
Kulkarni and P. Mohite, "Analysis of Impact of Image
Restoration and Segmentation on Classification
Model," 2023 7th International Conference On
Computing, Communication, Control And Automation
(ICCUBEA), Pune, India, 2023, pp. 1-5, doi:
10.1109/ICCUBEA58933.2023.10392033.
Laura A. Martinho, João M. B. Calvalcanti, José L. S. Pio
and Felipe G. Oliveira “Diving into Clarity: Restoring
Underwater Images using Deep Learning,” Journal of
Intelligent & Robotic Systems, 110, 32 (2024).
A. Nikitha, J. Geetha and D. S. JayaLakshmi, "Handwritten
Text Recognition using Deep Learning," 2020
International Conference on Recent Trends on
Electronics, Information, Communication &
Technology (RTEICT), Bangalore, India, 2020, pp.
388-392, doi: 10.1109/RTEICT49044.2020.9315679.
S. S. Harsha, B. P. N. M. Kumar, R. S. S. R. Battula, P. J.
Augustine, S. Sudha and T. Divya., "Text Recognition
from Images using a Deep Learning Model," 2022 Sixth
International Conference on I-SMAC (IoT in Social,
Mobile, Analytics and Cloud) (I-SMAC), Dharan,
Nepal, 2022, pp. 926-931, doi: 10.1109/I-
SMAC55078.2022.9987404.
A. Shrivastava, J. Amudha, D. Gupta and K. Sharma, "Deep
Learning Model for Text Recognition in Images," 2019
10th International Conference on Computing,
Communication and Networking Technologies
(ICCCNT), Kanpur, India, 2019, pp. 1-6, doi:
10.1109/ICCCNT45670.2019.8944593.
A. Ansari, B. Kaur, M. Rakhra, A. Singh and D. Singh,
"Handwritten Text Recognition using Deep Learning
Algorithms," 2022 4th International Conference on
Artificial Intelligence and Speech Technology (AIST),
Delhi, India, 2022, pp. 1-6, doi:
10.1109/AIST55798.2022.10065348.
A. Thilagavathy, K. H. Suresh, K. T. Chowdary, M. Tejash
and V. L. Chakradhar, "Text Detection based on Deep
Learning," 2023 International Conference on
Innovative Data Communication Technologies and
Application (ICIDCA), Uttarakhand, India, 2023, pp. 1-
6, doi: 10.1109/ICIDCA56705.2023.10099672.
S. MS, S. G and A. D. R, "Handwritten Text Recognition
Using Machine Learning and Deep Learning," 2023
Eighth International Conference on Science
Technology Engineering and Mathematics
(ICONSTEM), Chennai, India, 2023, pp. 1-4, doi:
10.1109/ICONSTEM56934.2023.10142716.
R. Sharma, B. Kaushik and N. Gondhi, "Character
Recognition using Machine Learning and Deep
Learning - A Survey," 2020 International Conference
on Emerging Smart Computing and Informatics
(ESCI), Pune, India, 2020, pp. 341-345, doi:
10.1109/ESCI48226.2020.9167649.
N. J and S. K. A M, "A Novel Text Recognition Using Deep
Learning Technique," 2024 International Conference
on Advances in Data Engineering and Intelligent
Computing Systems (ADICS), Chennai, India, 2024, pp.
1-6, doi: 10.1109/ADICS58448.2024.10533457.
R. Vaidya, D. Trivedi, S. Satra and P. M. Pimpale,
"Handwritten Character Recognition Using Deep-
Learning," 2018 Second International Conference on
Inventive Communication and Computational
Technologies (ICICCT), Coimbatore, India, 2018, pp.
772-775, doi: 10.1109/ICICCT.2018.8473291.
N. Teslya and S. Mohammed, "Deep Learning for
Handwriting Text Recognition: Existing Approaches
and Challenges," 2022 31st Conference of Open
Innovations Association (FRUCT), Helsinki, Finland,
2022, pp. 339-346, doi:
10.23919/FRUCT54823.2022.9770912.
INCOFT 2025 - International Conference on Futuristic Technology
714