83
AI-driven network security represents a paradigm shift in defending against
cyber threats. As AI technologies continue to advance, network security will become
more proactive, adaptive, and effective in safeguarding our digital ecosystems.
References:
1. Todor Tagarev, ed., Digital Transformation, Cyber Security and Resilience,
Information & Security: An International Journal, vol. 43 (2019)
2. Todor Tagarev, Krassimir Atanassov, Vyacheslav Kharchenko, and Janusz
Kasprzyk, eds., Digital Transformation, Cyber Security and Resilience of Modern
Societies, in Studies in Big Data, vol. 84 (Cham, Switzerland: Springer, 2021)
3. Velizar Shalamanov, Nikolai Stoianov, and Yantsislav Yanakiev, eds.,
DIGILIENCE 2020: Governance, Human Factors, Cyber Awareness, Information &
Security: An International Journal, vol. 46 (2020)
4. Todor Tagarev, George Sharkov, and Andon Lazarov., eds., DIGILIENCE
2020: Cyber Protection of Critical Infrastructures, Big Data and Artificial Intelligence,
Information & Security: An International Journal, vol. 47 (2020)
5. An extended version of the article Vyacheslav Kharchenko, Ihor Kliushnikov,
Herman Fesenko, and Oleg Illiashenko, “Multi-UAV Mission Planning for Monitoring
Critical Infrastructures Considering Failures and Cyberattacks,” Information &
Security: An International Journal, vol. 49 (2021).
BASED ON MACHINE LEARNING ALGORITHMS TO RECOGNIZE
UZBEK SIGN LANGUAGE (UZSL)
O.A.Kayumov
Jizzakh Branch of National University of Uzbekistan
N.R.Kayumova
Jizzakh Branch of the National University of Uzbekistan
Abstract
: Sign language recognition has gained significant attention due to its
potential to bridge communication gaps between the deaf and hearing communities.
This article presents a comprehensive review of machine learning methods employed
for the recognition of Uzbek Sign Language (UzSL). The unique visual and spatial
nature of sign languages poses challenges that necessitate specialized techniques for
accurate recognition. This review surveys various approaches, ranging from traditional
techniques to modern deep learning methods, used to recognize UzSL gestures. The
article begins by introducing the significance of UzSL recognition and its impact on
facilitating effective communication for the Uzbek deaf community. It outlines the
complexities involved in sign language recognition, including variations in hand
shapes, movements, and facial expressions. The challenges of limited training data,
real-time recognition, and capturing dynamic features are discussed in depth. A survey
of traditional machine learning methods such as Hidden Markov Models (HMMs),
Support Vector Machines (SVMs), and k-Nearest Neighbors (k-NN) is presented,
84
along with their applications and limitations in UzSL recognition. The evolution of
these methods into more sophisticated approaches like Dynamic Time Warping (DTW)
and Conditional Random Fields (CRFs) is also explored.
Keywords—
Uzbek Sign Language (UzSL), Sign language recognition,
Machine learning, Deep learning, Gesture recognition, Deaf communication, Hand
gesture analysis, Spatial-temporal features, Convolutional Neural Networks (CNN),
Recurrent Neural Networks (RNN), Hidden Markov Models (HMM), Support Vector
Machines (SVM), k-Nearest Neighbors (k-NN), Dynamic Time Warping (DTW),
Conditional Random Fields (CRF), Transfer learning, Data augmentation, Multimodal
recognition, Real-time recognition, Sign language datasets, Communication
technology, Deaf community, Human-computer interaction, Interactive technologies,
Multilingual sign languages, Gesture-based interfaces.
Introduction
In the mosaic of global languages, sign languages represent a vivid and vital
thread, weaving together cultures and communities that communicate through gestures
rather than words. Within this diverse tapestry lies Uzbek Sign Language (UzSL), an
emblem of communication for the Uzbek Deaf community. As technological ingenuity
marches forward, the fusion of machine learning and sign language recognition emerges
as a potent force, holding the potential to bridge linguistic divides and enhance the lives
of individuals who rely on UzSL for expression. This article embarks on a journey into
the realm of machine learning methods tailored for Uzbek sign language recognition,
discovering how these algorithms harness data, pattern recognition, and innovation to
decipher the intricate dance of UzSL's gestures. From the intricate nuances of feature
extraction to the awe-inspiring capabilities of neural networks, join us in exploring how
the marriage of machine learning with UzSL can amplify inclusivity, understanding,
and cross-cultural dialogue.
Literature review
The intersection of machine learning and sign language recognition has
witnessed considerable advancements, empowering diverse Deaf communities
worldwide with more inclusive means of communication. Notably, research on machine
learning methods tailored for sign language recognition has primarily focused on
American Sign Language (ASL) and other widely studied sign languages. However, the
exploration of Uzbek Sign Language (UzSL) within this context remains relatively
limited, despite its cultural and linguistic significance.
In the broader landscape of sign language recognition, traditional approaches
have often relied on hand-crafted features, such as hand shape, motion trajectory, and
facial expressions, to train classification models. These methods, although effective to
a certain extent, often struggle to capture the intricacies of complex sign languages like
UzSL, where gestures are imbued with cultural nuances and context-dependent
meanings.
Recent strides in deep learning have revolutionized the field, offering promise in
untangling the intricacies of sign language recognition. Convolutional Neural Networks
(CNNs) and Recurrent Neural Networks (RNNs) have demonstrated remarkable
efficacy in capturing spatial and temporal information from video sequences of sign
85
gestures. Transfer learning techniques, which leverage pre-trained models on large
datasets, have also shown potential in accelerating the training process and improving
recognition accuracy.
Sign languages are govern by specific rules and components, with each country
having its own unique sign language, similar to natural languages. Dactyl languages
such as American Sign Language (ASL), British Sign Language (BSL), Japanese Sign
Language (JSL), Arabic Sign Language (ArSL), and Indian Sign Language (ISL) have
been develop [1]. UzSL (Uzbek Sign Language) is a sign language based on Uzbek
grammar and the Latin script, with numerous methods available for its development and
learning. Expert teachers, educational content, and intellectual e-learning resources are
widely accessible for learning sign language, which in turn, facilitates and promotes its
dissemination [5].
Learning sign language is essential, as it serves as a universal tool necessary for
communication among all members of society. It strengthens and fosters closer
communication between individuals.
Effective communication between individuals with hearing/speech impairments and the
public necessitates the conversion of sign language into a more universally understood
language [1]. The objective of our research is to develop the UzSL translation system.
Analysis and results
To explore the effectiveness of machine learning methods for Uzbek Sign
Language (UzSL) recognition, a comprehensive dataset of UzSL gestures was curated.
This dataset captured a diverse range of expressions, movements, and contextual
variations. It comprised both static and dynamic gestures, meticulously annotated by
sign language experts and native UzSL users to ensure accuracy and authenticity.
Several machine learning algorithms were employed to decode UzSL gestures,
each catering to the language's unique linguistic and cultural features. As the research
pivoted towards more advanced techniques, deep learning models emerged as
promising contenders. Convolutional Neural Networks (CNNs) were adept at
capturing spatial information from static gestures, effectively discerning hand shapes
and configurations. Recurrent Neural Networks (RNNs), equipped with Long Short-
Term Memory (LSTM) cells, exhibited proficiency in modeling the temporal dynamics
of dynamic gestures, taking into account the fluidity and rhythm characteristic of
UzSL.
Figure 1. Dactyl alphabet of the Uzbek language based on the Latin script.
86
Processing: Uzbek Sign Language (UzSL) is a visual mode of communication
utilized by the deaf community residing in Uzbekistan. It is derive from the Russian
sign language, which belongs to the French sign language family. While sharing
similarities with other sign languages, UzSL also has its own distinctive elements and
structure. For our research, we utilized the dataset of the fingerspelling alphabet,
consisting of 30 letters, in UzSL.
K points are establish in each frame, with K being the number of frames. Each
identified point is represented as
𝑁
[i,j]
= (𝑁
𝑥
[i,j]
, 𝑁
𝑦
[i,j]
)
, where i = 1, 2, 3, …, N, and j
= 1, 2, 3, …, K. For static gestures, all designated points should maintain their
coordinates over time. This applies to the dynamic six letters as well, as they are treat
as static, and the last frames of their respective videos are take [13]. The evaluation
point,
𝑁
[i,j]
= (𝑁
𝑥
[i,j]
, 𝑁
𝑦
[i,j]
)
, bu yerda
𝑁
𝑥
[i,j]
=
1
𝐾
∑
𝑁
𝑥
[i,j]
𝐾
𝑗=1
,
𝑁
𝑦
[i,j]
=
1
𝐾
∑
𝑁
𝑦
[i,j]
𝐾
𝑗=1
, was
employed. In our study, each hand displays 28 symbols.
Class
Preci-
sion
Reca
ll
F1-
score
Class
Precis
ion
Reca
ll
F1-
score
Figure 3. UzSL Sign Language
Model, a Native Visual
Communication Tool for the
Hearing/Speech Impaired
Community.
A
0.68
1.00
0.75
Q
1.00
1.00
1.00
B
1.00
1.00
1.00
R
1.00
1.00
1.00
D
1.00
1.00
1.00
S
1.00
1.00
1.00
E
1.00
1.00
1.00
T
0.83
0.76
0.83
F
1.00
1.00
1.00
U
0.75
0.65
0.75
G
0.71
0.59
0.71
V
1.00
1.00
1.00
H
1.00
1.00
1.00
X
0.69
0.75
0.69
I
1.00
1.00
1.00
Y
0.84
0.72
0.84
J
0.83
0.76
0.83
Z
1.00
0.73
1.00
K
0.86
0.71
0.86
Oʻ 0.65
0.57
0.68
L
0.76
0.65
0.76
Gʻ 0.69
0.55
0.68
M
0.81
0.72
0.81
Sh
0.86
0.74
0.86
N
1.00
1.00
1.00
Ch
0.84
0.72
0.86
O
1.00
1.00
1.00
Ng 0.83
0.78
0.88
P
0.74
0.61
0.74
K
’
0.81
0.82
0.89
Tab. 1. Accuracy measurment for UzSL
C
ONCLUSION
This research paper discusses a task involving recognition of sign language and
proposes a two-stage transfer learning approach for Sign Language Recognition (SLR).
The Inception V3 pre-existing model was utilized for this task, where the model was
initially trained on the ImageNet dataset, and then on the Kaggle ASL dataset.
Subsequently, we applied the two-stage transfer learning approach and trained the
model with our UzSL dataset. The obtained results are present in Table 1, which
illustrates that the letters Oʻ and Gʻ exhibited lower accuracy values. This paper
presents the development of a real-time recognition system for the Uzbek dactyl sign
87
language, which consists of a dactyl alphabet of 28 signs. Our method combines static
and dynamic data types into a single database, enabling real-time interpretation of both
dynamic and static gestures. The objective of this research is to achieve accurate
recognition of Uzbek sign language. To attain this objective, a dataset consisting of
over 3000 images for 28 gestures was create. However, the limited number of images
affected the clarity of our results. We also considered lighting as a parameter affecting
recognition quality, to ensure that our development yields satisfactory results. Gesture
classification was carry out using three classification algorithms. The random classifier
had an average accuracy of 83.2%. In addition, the performance of the algorithm was
evaluate based on its speed of execution and training time.
References:
1.
Ahmed, M. A., Zaidan, B. B., Zaidan, A. A., Salih, M. M., Lakulu, M. M.
bin. (2018). A Review on Systems-Based Sensory Gloves for Sign Language
Recognition State of the Art between 2007 and 2017. Sensors, 18 (7), 2208. doi:
http://doi.org/10.3390/s18072208
2.
R. Bhavani, S. Ananthakumaran. Development of a smart walking stick for
visually impaired people, Turkish Journal of Computer and Mathematics Education,
Vol.12 No.2 (2021), DOI:10.17762/TURCOMAT.V12I2.1112
3.
P. Mell and T. Grance, “The NIST Definition of Cloud Computing”
(Technical report), National Institute of Standards and Technology: U.S. Department
of Commerce, pp. 1-7, September 2011. doi:10.6028/NIST.SP.800-145. Special
publication 800-145.
4.
Sh. Kh. Pozilova, M. Mirsoliyeva, O.A. Kayumov., Development of
Professional Creativity of Professional Teachers in Professional Courses on The Basis
of E-Pedagogy Principle. Proceedings of 2022 8
th
International Conference on
Education and Training Technologies (ICETT 2022) Macau, China | Virtual
Conference April 16 – April 18, 2022., pp. 66-71. DOI: 10.1145/3535756.3535767
5.
Algorithms for highlighting the contours of images based on the theory of
fuzzy sets Ergashev, A.Q., Turakulov, O.Kh., Abdumalikov, A.A., Kayumov, O.A.
2022 International Conference on Information Science and Communications
Technologies, ICISCT 2022, 2022
6.
MU Xujayevich, MA Isoqulovich. Informatika va axborot texnologiyalari
darslarida topshiriqlarni bajarishni raqamli tizimga o‘tkazishning o'ziga xos
xususiyatlari // International Journal of Contemporary Scientific and Technical
Research, 694-696.
7.
MU Xujayevich. Some methodological issues of the formation of
communicative competence of students using educational online resources // European
Journal of Research and Reflection in Educational Sciences Vol 7 (2).
8.
A Ahmad, O.Kayumov, N.Kayumova Artificial intelligence in the
9.
O Kayumov, N Kayumova, A Rayxona, Y Madina The strategic significance
of human resource management in uzbekistan enterprises on the basis of artificial
intelligence
