Enhancing Kurdish Sign Language Recognition Through Random Forest Classifier and Noise Reduction via Singular Value Decomposition (Svd)

Abstract

Deaf people around the world face difficulty communicating with others. Hence, they use their own language to communicate with each other. This paper introduces a new approach for Kurdish sign language recognition using the random forest classifier algorithm aiming to facilitate communication for deaf communities to communicate with others without relying on human interpreters. On the other side, for further enhancement of the images captured during recognition linear algebra techniques have been used such as singular value decomposition for image compression and Moore–Penrose inverse for blur removal. Kurdish language has 34 alphabets and (10 numeric numbers 10, . . . ,3 ,2 ,1). Additionally, three extra signs have been created and added to the dataset, such as space, backspace, and delete sentences for the purpose of real-time translation. A collection of 800 images has been gathered for each character, out of 800 images, only 80 per character were used due to their similar positions but varied alignment, totalling 3,520 images for the dataset (44 characters 80 images each). Two simulation scenarios were carried out: one with optimal conditions - a white background and adequate lighting, and another with challenges such as complex backgrounds and varied lighting angles. Both achieved high match rates of 96% and 87%, respectively. Further, a classification report analyzed precision, recall, and F1 score metrics.