Indian Sign Language Recognition using Vision Transformer based Convolutional Neural Network

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 10 Issue: 04 | Apr 2023 www.irjet.net p-ISSN: 2395-0072
© 2023, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 1154
Indian Sign Language Recognition using Vision Transformer based
Convolutional Neural Network
Sunil G. Deshmukh1, Shekhar M. Jagade2
1Department of Electronics and Computer Engineering, Maharashtra Institute of Technology, Aurangabad
431010, Maharashtra, India
2Department of Electronics and Telecommunication, Sri N B Navale Sinhgad College of Engineering, Solapur-
413255, Maharashtra, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Hand gesture recognition (HGR) is a popular
issue in the areas of learning algorithms and visual
recognition. Certain Human-Computer Interactions
technologies also need HGR. Traditional machine learning
techniques and intricateconvolutional neural networks(CNN)
have been employed to HGR up till now. Despite the fact that
these approaches work adequately well on HGR, we employed
a more modern model, vision transformer, in this research.
Vision Transformer (ViT) is created to enhance CNN's
performance. ViT has a strong similarity to CNN, but its
classification work utilizes distinct layers. The ViT was
performed to gesture datasets via learning algorithms. In
trials, a testing dataset having crossed test strategy is
evaluated, and classification accuracy is employed as
productivity metric. According to test findings, the presented
approach framework attains an achieved accuracy of 99.88%
on the image database used, which isconsiderablyhigherthan
the state-of-the-art. The ablationstudyalsosupportstheclaim
that the convolutional encoding increases accuracy on HGR.
Key Words: Convolutional neural network, Vision
transformer (ViT), Transfer learning, Training of
images, Accuracy, Hand gesture, Human-computer
interaction (HCI).
1. INTRODUCTION
Direct contact is becoming the most popular way for users
and machines to communicate. People connect with one
another naturally and intuitively through contactless
techniques including sound as well as body actions. The
versatility and efficacy of these contactless communication
techniques have motivated several researchers to adopting
them to further HCI. Gestures are a significant partofhuman
language and an essential non-contact communication
technique. Wearable data gloves were frequentlyusedinthe
past to grab the positions and angles of every user's jointsas
they moved.
The complexity and price of a worn sensor have limited the
extensive application of such a technology. Gesture
recognition refers to a computer's ability to understand
gestures and execute specified instructions in response to
such motions. The main goal of HGR is to establish a system
that can identify, analyze, and communicate information
based on specific motions [1].
Techniques for recognizing gestures on the basis of
contactless visual inspection are now prominent. This is
because their accessibility and price. Hand gestures are an
expressive communication technique employed in the
healthcare, entertainment, and educational sectors of the
economy, as well as to assist people with special needs and
the elderly. For the purpose of identifying hand gestures,
hand tracking which combines a number of computer vision
operations such as hand segmentation, detection, and
tracking is crucial. HGR are used to convey information or
emotions in sign language to those who have hearing loss.
The major problem is that the typical person may easily
misinterpret the message. AI and computer vision
developments may be utilized to identify and comprehend
sign language.
With the assistance of moderntechnology,thetypical person
may learn to recognize sign language. Thisarticleintroduces
a deep learning-based technique for hand gesture
identification. In this regard, operating the system remotely
necessitates the use of gestures. The devices record human
motions and recognize them as the ones that are used to
control them. The movements employ a variety of modes,
including static and dynamic modes. The static gestures are
maintained while the dynamicgesturesshifttovariousareas
when the machine is being controlled. So, rather than using
static gestures, it is essential to identify or recognize
dynamic motions. The camera that is connected to the
apparatus initially captures people's movements [2].
When the backdrop of any motions that were identified is
removed, the gesture's foregroundisgathered.Tolocateand
eliminate the noises in the foreground gesture, filtering
techniques are applied. These noise-removed gestures are
compared to pre-stored and taught movements in order to
verify the meaning of the gestures. The automotive and
consumer electronics sectors utilize a gesture-based
machine operating system that doesn't require any human
input. Static and dynamic gestures, as well as online and
offline actions, can all be classified as human gestures. The
machine's icons can be changed using offline gestures, but

the system or menu where the items are shown cannot be
altered. The movements of the internet cause the machine's
symbols to shift or tilt in a variety of ways.
Online gestures are significantly more useful than offline
gestures in real-time machine operating systems. These
strategies required a large number of training samples and
could not handle large training datasets. This flaw isfixed by
suggesting ViT-based CNN in this study. The complexity of
this technique is low, and it does not require a big amount of
data to train. The incorporation of a deep learningalgorithm
and a cutting-edge segmentation approach into a HGR
system is an original component of the proposed study.
Enhancing users to communicate by programming software
more receptive to user demands is the fundamental aim of
human-computer interaction(HCI).Gesturesareintentional,
suggestive body movements that include the physical action
of the hands, wrists, face, shoulders, and face with the
intention of engaging with the surroundings or conveying
the relevant information. The creation of human-centric
interfaces is made possible by hand gesture recognition
(HGR), a crucial job in HCI and a highly helpful method for
computers to comprehend human behavior.
A vital part of our everyday lives is being able to
communicate effectively. People who are deaf or dumb find
it challenging to engage with people due to their incapacity
to talk and listen. One of the most effective and well-known
techniques is arguably the use of hand gestures, sometimes
referred to as sign language. Programs that can identify sign
language motions and movements must be developed in
order for deaf and dumb people to communicatemoreeasily
with those who do not understand sign idioms. The purpose
of this study is to employ sign languages as a first step in
removing the barrier to communication between hearing-
impaired and deaf people. When employed to solve picture
recognition issues in computer vision, modern (CNN) yield
successful performance. The most effective way to build a
complete CNN network is via the use of transfer learning. In
a broad range of fields, including automation, machine
learning is used. Aspects of artificial intelligence include
techniques for measuring, recording, detecting, monitoring,
identifying, or diagnosing physical phenomena. Moreover, a
variety of sensors, such as vision-based sensors, capacitive
sensors, and motion-based sensors can record hand
movements [3].
Data Glove Techniques: Glove-based sensors are the only
technologies that can handle the complex needs of hand-
based input for HCI. In order to determine the hand
postures, this approach uses mechanical or optical sensors
that are mounted to a glove and translate finger flexionsinto
electrical impulses. The following disadvantages renderthis
innovation less well-liked: The user is required to carry a
burden of wires that are linked to the processor and it also
needs calibrating and setup processes, thus contact with the
computer-controlled surroundings lacks its ease and
naturalness.TechniquesthatarebasedonvisionApproaches
based on system vision have the ability to obtain more non-
intrusive, natural solutions since they are based on how
people interpretinformation abouttheirenvironment.While
it is challenging to create a vision-basedinterfaceforgeneral
use, it is possible to do so for a controlled environment
without facing many difficulties, such as accuracy and
processing speed [4].
Heap et al. [5] developed a deformable 3D hand model, and
they surface-mapped the whole hand using PCA from
training instances. Real-time tracking is made possible by
locating the most likely to be distorted model that matches
the picture. While more processing is necessary to extract
valuable higher metadata, such as pointing direction, it has
been shown that such a representation is quite successful at
precisely locating and tracking the hand in pictures.
Nevertheless, the approach is not scale and rotation
invariant and cannot tackle the obstruction issue.
Compositional methods are used in to identifyhand posture.
Configurations of parts serve as the foundation for a hand
posture depiction. Characteristics are classified in
accordance with the perceptual principles of grouping to
provide a list of potential candidate compositions. Based on
overlapped sub-domains, these subgroups provide a sparse
picture of the hand position.
A novel method for identifying static motions in challenging
circumstances based on wristband-based contour features
was proposed by Lee et al. [6]. Pair of black wristbands is
applied to both hands to properly divide the hand area. By
the use of a matching algorithm, thegestureclassisdetected.
When the background colour reappears, the system fails to
effectively segregate the hand region. For the purpose of
detecting static hand gestures, Chevtchenko et al. [7]
optimized a coalitation of characteristics and dimensions
using a multi-objective evolutionary algorithm. Up to 97.63
percent identification accuracy was reached on 36 gesture
postures from the MU dataset. Theaforementionedaccuracy
was attained utilizing a holdout cross-validationtestand the
combined Gabor filter and Zernike moment features. A
variety of geometriccharacteristicsincludingangle,distance,
and curvature features were also producedbythecontour of
the hand motion, and these aspects were classified as local
descriptors. Following that, local descriptorswereimproved
using the Fisher vector, and gesture recognition was
accomplished using a support vector machine classifier.
Deep features from AlexNet's fully connected layer and VGG
16 were used to recognize sign language. The obtained
attributes were classified using the SVM classifier. The
recognition performance was estimated to be 70% using a
conventional dataset and the leave-one-subject-out cross-
validation test. The results show that for RGB input photos,
detection performance is constrained by background
variation, human noise, and high inter-classsimilarityinASL
gesture postures.

Parvathy et al. [8], the authors deploy a vision-based HGR
system to extract key characteristics from radar-based
pictures before categorizing them with a 96.5% accuracy
SVM classifier. Deep learning models based on (CNN)
architecture have attracted more interest in human
computation and object detection. Convolutional filters are
used by CNN to extract the key details of an image's
subject the object classification convolution processes that
cover important attributes should therefore be minimized.
The authors Zhan et al. [9] have suggested a CNN model for
the real-time HGR. This model's accuracy was 98.76% using
a dataset of 500 photos and 9 distinct hand motions. They
also suggested a double dimensional (2D) CNN model for
dynamic HGR that includes of max and min resolution sub-
networks, which was later demonstratedtohaveattainedan
accuracy of 98.2% on the same dataset.
For dynamic HGR that incorporatesactionsaswell asmotion
sensing, the suggested works are not appropriate. So, it is
necessary to investigate the selection of reliable classifiers
and suitable hyper-parameters for the same. Adithya and
Rajesh [10] suggest a deep CNN architecture for HGR. The
suggested architecture avoids detecting and segmenting
hands in collected pictures. Theauthorstestedthesuggested
architecture on the NUS hand posture and American Finger
spelling A datasets, achieving accuracy of 94.26% and
99.96%, respectively. TheauthorsIslametal.[11]developed
a CNN model with data augmentation for static HGR. In this
case, 8000 photographs from the dataset are utilized for
training, whereas 1600 images are used for testing after
being divided into 10 classes. The dataset has been
augmented using various data augmentations such as
zooming, re-scaling, rotation, shearing, height, and width
shifting.
Neethu et al. [12] used a cutting-edge deep CNN model to
perform hand motion recognition and HGR. Here, the hand
and finger photos are separated into separate mask images,
which are subsequently sent to CNN for classification into
several groups. The results demonstrate that the suggested
CNN model produced a 90.7% recognition accuracy. Wu
[13], suggested a double channel CNN (DC-CNN) model for
HGR. The hand gesture photos and hand edge images are
created here by pre-processing the source photographs. The
CNN channels then categoriesthesepicturesusingtwofeeds.
These pictures are then sent into two CNN channels for
classification.Thesuggested DC-CNN demonstratedaccuracy
of 98.02% and 97.29% on the Jochen-Triesch Database and
the NAO Camera hand posture Database, respectively. The
suggested DC-CNN approach has not been tested with
complicated backdrop pictures. Further features are being
added to make the model more versatile. The model's
applicability to dynamic HGR must be investigated.
Lai and Yanushkevich [14] proposed a mix of CNN and
recurrent neural network (RNN)forautonomousHGRbased
on depth and skeleton data. RNN is utilized to process the
skeleton data in this case, whereas CNN is used to analyze
the depth data. Using the dynamic hand gesture-14/28
dataset, the suggested model attained an accuracy of
85.46%. The presented approach may be expanded to
recognize human activities, which is an intriguing subject to
investigate. Bao et al. [15] suggested a deep CNN model for
small HGR directly from pictures without segmentation or
pre-processing. The suggested HGR network candistinguish
between up to seven groups. The suggested model attained
accuracy of 97.1% for simple backdrop and 85.3% for
complicated background, respectively. It has been said that
the proposed solution has the benefit of being able to be
implemented in real-time
A realistic and effective way for creating the model's
contours which are then evaluated with the picture data is
produced by using quadrics to create the 3D model. An
Unscented Kalman filter (UKF), which reduces the
geometrical uncertainty between the profiles and edges
retrieved from the pictures, is used to predict the posture of
the hand model. Although giving more accuracy than the
enhanced Kalman filter, the UKFallowsforfasterframerates
than more complex estimating techniques like particle
filtering. The hand gesture recognition job may depend on a
variety of input data types and combinations, much like
other computer vision-related tasks. As a result,
methodologies described in the literature may generally be
divided into two categories: monotonic and multifunctional.
To address the action and sign language recognition
challenges, the authors of suggest transformer-based
methods that are comparable to ours. An input translator
and recognizing paradigm that resembles the Faster R-
structure CNN's uses a slightly adapted variant of the
transformer architecture. A data extractor and a
transformer-like framework are combined to recognize
actions in real-time. The temporal linkages are not explicitly
modelled since it uses 1D convolutional layers between
consecutive decoder blocks rather than any kind of
positional encoding. On the contrary side, in our method,
positional encodings is used to encapsulate the temporal
features on the frames order (PE) [16].
The CNNs is the state-of-the-art in machine learning and
extensively employed for various image identification
applications, have a strong competitor in the shape of Vision
Transformer (ViT). In terms of accuracy and computational
efficiency, ViT models performed approximately four times
better than the most advanced CNNs currently available.
New vision transformer frameworks have also exhibited
surprising skills, reaching equivalent and even superior
performance than CNNs on several computer vision
applications, despite convolutional neural networks having
long since dominating the area of machine learning.
Here, it has been researched and developed a Transformer-
based framework to recognize hand gestures in this article.
In a sense, we take use of the Transformer architecture's

recent ground-breaking work in solving many complicated
ML issues, as well as its enormous potential to use more
input parallel processing with attentive mechanisms. To
solve the issues mentionedabove,VisionTransformers(ViT)
might be regarded as an acceptable architecture. The
presented Vision Transformer-based Hand Gesture
Recognizing (ViTHGR) framework may address issues with
assessment accuracy and training time that are often
encountered when using other networks like the traditional
ML techniques. As a result, the primary signal is divided into
smaller sections using a certain window opening in the
recommended ViT-HGR topology, andoneofthesesegments
is then sent to the ViT for additional assessment.
2. VISION TRANSFORMER (ViT)
The advent of Vision Transformer (ViT) puts CNN, the state-
of-the-art in computer vision that is so often used in a
variety of image identification applications, into intense
competition. Convolutional neural networks (CNNs) cannot
compete with the ViT models in terms of processing power,
effectiveness, and accuracy.Modernperformancecriteriafor
transformer topologies in the area of natural language
processing.
In certain computer vision tasks, the transformer
outperforms existing convolution approaches [10]. A vision
transformer is a transformer usedspecificallyfora computer
vision job (ViT). ViTs perform impressively and with great
promise in computer vision applications. ViT offers two key
advantages: 1) Self-attention method, in which the model
interprets a wide variety of input seeds (tokens) in an all-
encompassing context. 2) The capacity to practice
challenging activities. The ViT concept is shown in Fig. 1,
where initial photos are patched together in line with the
prototype model. Patches are then sent immediatelyinto the
layer of the straight projections. The procedure of patch
embedding is carried out in the second step. The sequencing
of the inserted patches now includes the class token. As a
result, patches grow in size by one. Positional embedding
additionally adds embedded patches to the database spatial
sequencing of patches. Lastly,theencoderlayerservesasthe
first transformer layer and is given patch embedding and
positioning of encoding with a class token. The most crucial
part of a transformer, notably in ViT, is encoding. The
feeding forward layer receives the output of the attentive
layer, which then produces the encoder'sfinal outcome[17].
Fig-1: Sketch of overall ViT concept
The ViT encoder has many stages, and each component
consists of three significant processing components:
1. Layer Norm
2. Multi-head Attention Network (MSP)
3. Multi-Layer Perceptions (MLP)
1. Layer Norm helps the model adjust to the
differences between the training pictures while
keeping the training phase on track.
2. A network called the Multi-head AttentionNetwork
(MSP) creates attention mappings from the
integrated visual signals that are provided. These
concentration maps assist the network in
concentrating on the image's most crucial areas,
such as object (s). The idea of attention mapping is
similar to that explored in the research on
conventional computer recognition (e.g., saliency
maps and alpha-matting).
3. The Gaussian Errors Linear Unit (GELU) is the last
layer of the two-layer classification network known
as MLP. The final MLP block, commonly known as
the MLP head, serves as the transformer's output.
Data augmentation may be used on this output to
provide categorization labels.
The transformer's architectural has been preserved,andthe
majority of the alterations relate to how the photos are
altered before being sent to the transformer.WhereasVision
Transformer (ViT) uses pictures as its input, NLP-based
models take a sequence of words as input. Images are split
into 2D patches of a given size to account for this disparity
(usually 16x16). Prior to the patches being supplied to the
Transformer as a flattened 1D sequence, recommends
utilizing linear projections on flattened 2D patches and
positional embedding to get patch embedding while

retaining positional information (just like tokens). The
Transformer Encoder, which is composed of an attention
layer and a Multi-Layer-Perceptron, receives a series of
patches as input in the basic model's architecture (MLP).
This is then sent to a MLP that is not connected to the
encoder, and this MLP eventually produces a probability
distribution of all the classes. The picture is initially run
through a featured extruder, which identifies crucial
characteristics that describe the image and normalizes
images across RGB channels while maintaining mean and
standard deviations. The ViT model is then applied to the
picture, with each layer teaching the model that something
new before producing the result [17].
Vaswani et al. [18] served as an inspiration for the
Transformers encoding image. A picture is split into small
patches, each of which is linearly embedded, positional
embeddings are added, and the resultant sequences of
matrices are sent to a typical Transformer encoding. It has
been used the conventional method for adding an additional
trainable "classifying token" to the sequences as a way to do
classification.
3. RESULTS AND DISCUSSION
In this article, the interpretation of predefined Indian sign
languages is examined. The suggested strategy may aid
disabled individuals in conversing with normal individuals.
In the perspective of the Indian language,thedatasetutilized
for research is a static ISL database with digit andalphabetic
signals. Functioning with the transformer encodingdoesnot
need any data preparation. ViT assists the model in
enhancing its performance. The performance of the
suggested technique is shown in Table 1. Recall is the total
proportion of entries that were mistakenly categorized as
negative, calculated by (“equation2”)dividingthenumberof
real positives by the summation of true positives and false
negatives. Precision and F-1 score are calculated by
“equation 1” and “equation 3”, respectively.
Mathematically, the report can be represented in a similar
manner as;
(1)
True positive
precision
True positive False positive


Re (2)
True positive
call
True positive Negative


*Re
1- 2* (3)
Re
precision call
F Score
precision call


Table -1: Performance of vit methodology
As seen in Fig. 2. the proposed model demonstrates great
convergence throughout the training and validation phases.
The samples were evaluated by means of the ViT model.
These are the testing accuracy results for ten epochs. The
total average accuracy was 99.88%; as showninFig.2.outof
the 48281 images utilized for testing, 3531 were correctly
categorized, while 165 were misclassified. It is seen that the
ViT model performed having good index of accuracy level.
From Fig. 3 it can be observed that the convergencetimeand
losses are minimized. Table 2 displays the training loss,
percentage accuracy, validationaccuracy,andvalidationloss
as they occurred with the response time.
Fig-2: Model accuracy of ViT model
A confusion matrix is also plotted in Fig. 4. A confusion
matrix displays the real positives and negatives identified
during the execution of the algorithm. This illustrateswhere
the algorithm went wrong as well as the number of accurate
exclusions and false positives. The genuinelabelingisshown
on the y axis and the anticipated labeling is displayed on the
x axis. The confusion matrix helps to determine whether or
not the bulk of the model's predictions are true.
Data
set
Accuracy Precision F1-
Score
Classification
Error
ISL 99.88% 0.99 0.99 0.72

Fig-3: Model loss of ViT model
Table-2: Percentage accuracy, validation accuracy and
validation loss of vit model
Epoch
Res.
Time
(S/step)
T-Loss
%
Accu.
Val.
Loss
Val.
Accu.
1/10 871s 18m 1.5406 0.6672 0.3923 0.9912
2/10 911s 19m 0.4262 0.9386 0.1380 0.9939
3/10 891s 18m 0.2336 0.9715 0.0729 0.9955
4/10 864s 18m 0.1587 0.9820 0.0483 0.9972
5/10 854s 18m 0.1200 0.9865 0.0341 0.9980
6/10 858s 18m 0.0962 0.9899 0.0263 0.9981
7/10 852s 18m 0.0804 0.9920 0.0208 0.9985
8/10 853s 18m 0.0701 0.9929 0.0177 0.9939
9/10 892s 18m 0.0604 0.9942 0.0147 0.9985
10/10 908s 19m 0.0531 0.9948 0.0122 0.9988
Res. Time = Response time, T-Loss: Training loss, % Accu.:
Percentage Accuracy, Val. Loss: Validation Loss, Val. Accu.:
Validation accuracy.
Fig-4: Confusion matrix of ViT model
4. CONCLUSION
In this research, the viability of the ViT approach for the
categorization of the HGR is investigated. The shortcomings
of the Convolutional networks were addressed using the
newly introduced strategy known asViT.Thesuggested ViT-
HGR system can categorize a significant variety of hand
gestures correctly from start without the necessity of data
augmenting and/or transfer learning, hence overcomingthe
retraining time issueswithrecurrentplatforms.Accordingto
test findings, the presented approach framework attains an
achieved accuracy of 99.88% on the image database used,
which is considerably higher than the state-of-the-art. The
ablation study also supports the claimthattheconvolutional
encoding increases accuracy on HGR. In next research, more
pre-trained ViT algorithms will be utilizedforthe purpose of
improving the accuracy of hand gesture identification.
Furthermore, additional hand gesture datasetsaswell asthe
MLP mixing models will be employed in future
investigations.
REFERENCES
[1] Z. Ren, J. Yuan, J. Meng, and Z. Zhang, “Robustpart-based
hand gesture recognition using kinect sensor,” IEEE
transactions on multimedia, vol. 15, no. 5, 2013, pp.
1110–1120.
[2] S. Ahmed, K. D. Kallu, S. Ahmed, and S. H. Cho, “Hand
gestures recognition using radar sensors for human-
computer-interaction: A review,” Remote Sens., vol.
13,no. 3, Feb. 2021.
[3] Van den Bergh, M., & Van Gool, L, “Combining RGB and
ToF cameras for real-time 3D hand gesture interaction,”

In 2011 IEEE workshop on applications of computer
vision (WACV), January 2011, pp. 66-72,
[4] Karbasi, M.; Bhatti, Z.; Nooralishahi, P.; Shah, A.;
Mazloomnezhad, S.M.R, “Real-time hands detection in
depth image by using distance with Kinect camera,” Int.
J. Internet Things, Vol. 4, 2015, pp. 1–6.
[5] A. J. Heap, D. C.Hogg, “Towards 3-D hand tracking using
adeformable model,” In 2nd International Face and
Gesture Recognition Conference (1996), pp 140–45.
[6] D. Lee and W. You, “Recognition of complex static hand
gestures by using the wristband‐based contour
features,” IET Image Processing, vol. 12, no. 1, 2018, pp.
80–87.
[7] S. F. Chevtchenko, R. F. Vale, and V. Macario, “Multi-
objective optimization for hand posture recognition,”
Expert Systems with Applications, vol. 92, 2018, pp.
170–181.
[8] Parvathy P, Subramaniam K, Prasanna Venkatesan G,
Karthikaikumar P, Varghese J, Jayasankar T,
“Development of hand gesture recognitionsystemusing
machine learning,” J Ambient Intell Humaniz Comput
12(6): 2021,6793–6800.
[9] Zhan F, “Hand gesture recognition with convolution
neural networks,” In: 2019 IEEE 20th international
conference on information reuse and integration for
data science (IRI), 2019, pp 295–298.
BIOGRAPHIES
Sunil G. Deshmukh acquired the
B.E. degree in Electronics
Engineering from Dr. Babasaheb
AmbedkarMarathwada University,
Aurangabad in 1991 and M.E.
degree in Electronics Engineering
from Dr. Babasaheb Ambedkar
Marathwada University (BAMU),
Aurangabad in 2007. Currently, he
is pursuing Ph.D. from Dr.
BabasahebAmbedkarMarathwada
University, Aurangabad, India. He
is the professional member of
ISTE, His areas of interests include
image processing, Human-
computer interaction.
[11] Islam MZ, Hossain MS, ul Islam R, Andersson K,
“Static hand gesture recognition using convolutional
neural network with data augmentation,”. In:2019joint
8th international conference on informatics, electronics
vision (ICIEV) and 2019 3rd international conferenceon
imaging, vision pattern recognition (icIVPR), 2019, pp.
324–329.
[10] Adithya V, Rajesh R, “A deep convolutional neural
network approach for static hand gesture recognition,”
Procedia Computer Science third international
conference on computing and network communications
(CoCoNet’19) 171, 2020, pp. 2353–2361.
[12] Neethu PS, Suguna R, Sathish D, “An efficient
method for human hand gesture detection and
recognition using deep learning convolutional neural
networks,” Soft Comput, vol. 24(20), 2020, pp. 15239–
15248,.
[13] Wu XY, “A hand gesture recognition algorithm based on
dc-cnn”, Multimed Tools Appl, vol. 79(13), 2020, pp.
9193–9205.
[14] Lai K, Yanushkevich SN, “Cnn+rnn depth and
skeleton based dynamic hand gesture recognition,” In:
2018 24th international conference on pattern
recognition (ICPR), 2018, pp 3451–3456.
[15] Bao P, Maqueda AI, del-Blanco CR, García N, “Tiny
hand gesture recognition without localizationvia a deep
convolutional network,”.IEEETransConsumerElectron,
vol. 63(3), 2017, pp. 251–257.
[16] Song, L.; Hu, R.M.; Zhang, H.; Xiao, Y.L.; Gong, L.Y,
“Real-time 3d hand gesture detection from depth
images”, Adv. Mater. Res., Vol. 756, 2013, pp. 4138–
4142.
[17] Cheng, P. M., and Malhi, H. S. “Transfer learning with
convolutional neural networks for classification of
abdominal ultrasound images,” Journal of digital
imaging, Springer, Vol. 30(2), 2017, pp.234-243.
[18] Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J.,
Jones, L., Gomez, A.N., Kaiser, Ł. and Polosukhin, I.
“Attention is all you need”, In: 31st Conference on
Neural Information Processing Systems (NIPS 2017),
Long Beach, CA, USA, 2017, pp. 1-15.

Shekhar M. Jagade secured the
B.E. degree in Electronics and
Telecommunication Engineering
from Government Engineering
College, Aurangabad in 1990 and
M.E. degree in Electronics and
Communication Engineering from
Swami Ramanand Teerth
Marathwada University,Nandedin
1999 and Ph.D from Swami
Ramanand Teerth Marathwada
University, Nanded in2008.Hehas
published nine paper in
international journals and three
papers in international
conferences. He is working as a
vice-principal in the N. B. Navale
Sinhgad College of Engineering,
Kegaon-Solapur, Maharashtra,
India. He is the professional
member of ISTE, ComputerSociety
of India, and Institution of
Engineers India. His field of
interests includes image
processing, micro-electronics,
Human-computer interaction.

Indian Sign Language Recognition using Vision Transformer based Convolutional Neural Network

More Related Content

Similar to Indian Sign Language Recognition using Vision Transformer based Convolutional Neural Network (20)

More from IRJET Journal (20)

Recently uploaded (20)

Indian Sign Language Recognition using Vision Transformer based Convolutional Neural Network