Scholarlink

<p>Communication has long relied on hand gestures, and sign language serves as a visual means of interaction. Sign Language Recognition (SLR) plays a vital role in bridging communication between hearing-impaired and deaf individuals. However, there is no universal sign language, as different countries use their own systems, such as American Sign Language, Korean Sign Language, and Japanese Sign Language. This manuscript proposes the development of a vision-based system for interpreting Indian Sign Language (ISL), optimized using a Physics-Informed Kernel Function Neural Network (DVS-ISL-PIKFNN). Initially, the input image is collected from the ISL Dataset, and pre-processed using the Fast Resampled Iterative Filter (FRIF) to enhance data quality, reduce image size and resolution, and minimize computational complexity. The pre-processed images are fed into the Random Quantum Circuits Transform (RQCT) for feature extraction, which is used to extract geometric features such as area, Irregularity Index, Solidity, and Equivalent Diameter. These features are subsequently fed into the PIKFNN for classification into fingerspelling and isolated word categories, while the Improved Chimp Optimization Algorithm (ICOA) is employed to optimize model parameters for enhanced performance. The proposed DVS-ISL-PIKFNN model is implemented in Python and evaluated using standard performance metrics, including accuracy, precision, recall, F1-score, and loss. Simulation results demonstrate that the proposed method achieves an accuracy of 99.6% for fingerspelling and 98.2% for isolated words, outperforming existing methods such as Vision-based hand gesture recognition using deep learning for the interpretation of sign language (VHGR-ISL-GCNN), Hand Gesture Recognition for Multi-Culture Sign Language Using General Deep Learning Network (HGR-MCSL-GCN) and CNN-based feature extraction and classification for sign language (FEC-SL-CNN), respectively. Furthermore, the integration of physics-informed constraints enables inherent robustness to geometric variations, such as rotation and scaling, reducing reliance on conventional data augmentation techniques and improving generalization capability.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An advanced model for gesture recognition in Indian sign languages classification-based accuracy improvement demonstrating robustness to rotation and scaling using deep learning approach

  • 期刊论文

  • 发表日期：

    2026年5月7日

查看全文
  • K. Priyadharshini,
  • Ajanthaa Lakkshmanan

摘要

Communication has long relied on hand gestures, and sign language serves as a visual means of interaction. Sign Language Recognition (SLR) plays a vital role in bridging communication between hearing-impaired and deaf individuals. However, there is no universal sign language, as different countries use their own systems, such as American Sign Language, Korean Sign Language, and Japanese Sign Language. This manuscript proposes the development of a vision-based system for interpreting Indian Sign Language (ISL), optimized using a Physics-Informed Kernel Function Neural Network (DVS-ISL-PIKFNN). Initially, the input image is collected from the ISL Dataset, and pre-processed using the Fast Resampled Iterative Filter (FRIF) to enhance data quality, reduce image size and resolution, and minimize computational complexity. The pre-processed images are fed into the Random Quantum Circuits Transform (RQCT) for feature extraction, which is used to extract geometric features such as area, Irregularity Index, Solidity, and Equivalent Diameter. These features are subsequently fed into the PIKFNN for classification into fingerspelling and isolated word categories, while the Improved Chimp Optimization Algorithm (ICOA) is employed to optimize model parameters for enhanced performance. The proposed DVS-ISL-PIKFNN model is implemented in Python and evaluated using standard performance metrics, including accuracy, precision, recall, F1-score, and loss. Simulation results demonstrate that the proposed method achieves an accuracy of 99.6% for fingerspelling and 98.2% for isolated words, outperforming existing methods such as Vision-based hand gesture recognition using deep learning for the interpretation of sign language (VHGR-ISL-GCNN), Hand Gesture Recognition for Multi-Culture Sign Language Using General Deep Learning Network (HGR-MCSL-GCN) and CNN-based feature extraction and classification for sign language (FEC-SL-CNN), respectively. Furthermore, the integration of physics-informed constraints enables inherent robustness to geometric variations, such as rotation and scaling, reducing reliance on conventional data augmentation techniques and improving generalization capability.