Title: Efficient Indian sign language recognition and classification using enhanced machine learning approach

Authors: Edwin Shalom Soji; T. Kamalakannan

Addresses: Department of Computer Science, Vels Institute of Science, Technology & Advanced Studies (VISTAS), Chennai, Tamil Nadu, India ' Department of Computer Science, Vels Institute of Science, Technology & Advanced Studies (VISTAS), Chennai, Tamil Nadu, India

Abstract: Deafness and voice impairment are two significant disabilities that make it difficult for people to communicate in verbal languages with others in a verbally communicating population. To solve this problem, the sign language recognition (SLR) system was constructed by combining machine learning and deep learning. The SLR employs hand gestures to convey messages. Earlier research aims to develop vision-based recognisers by extracting feature descriptors from gesture photos. When dealing with a large sign vocabulary recorded under chaotic and complex backgrounds, these strategies are ineffective. Hence, an improved convolution neural network is proposed in this paper to predict the most frequently used gestures in the Indian population with improved efficiency. The presented system is compared to SVM and CNN. The suggested approach is tested on 2,565 UCI instances and 22 training attributes. It showed both-handed ISL movements against various backgrounds. The augmented CNN has a precision of 89% and 90.1% accuracy, which is higher than most other approaches. According to this survey, we had an 83% recall and a 0.4 F score. Python evaluates our work.

Keywords: sign language recognition; SLR; machine learning; convolution neural network; CNN; Indian sign languages; ISLs; accuracy; precision.

DOI: 10.1504/IJCIS.2024.137405

International Journal of Critical Infrastructures, 2024 Vol.20 No.2, pp.125 - 138

Received: 16 Jun 2022
Accepted: 09 Aug 2022
Published online: 18 Mar 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article