Enhancing Sign Language Recognition through Deep CNN and Handcrafted Features

The current study presents a novel methodology for improving the recognition of Indian Sign Language (ISL) by combining Deep Convolutional Neural Networks (CNN) with manually designed features. The identification of ISL presents distinct difficulties as a result of its varied and subtle motions, requiring a strong and precise recognition system. To tackle this issue, we utilise the capabilities of deep learning with Convolutional Neural Networks (CNNs) to autonomously acquire distinctive characteristics from unprocessed data. Furthermore, we include meticulously constructed elements that are specifically tailored for ISL, effectively capturing fundamental attributes of hand gestures. Our methodology entails a multi-stage procedure, in which the deep Convolutional Neural Network (CNN) collects sophisticated characteristics from unprocessed ISL images, while the manually designed features offer additional information to enhance the recognition process. We showcase empirical findings on an extensive ISL dataset, showcasing the efficacy of our suggested methodology in enhancing recognition precision and resilience in contrast to conventional techniques. In addition, we do thorough assessments and comparative examinations to demonstrate the benefits of combining deep Convolutional Neural Networks (CNNs) with manually designed features for ISL recognition. The results of our study demonstrate that our method has the ability to greatly increase the accessibility and usability of ISL recognition systems.