Novel Method for Recognizing Sign Language using Regularized Extreme Learning Machine

Abstract

The ability of humans to effectively communicate through the use of hand signs has a wide range of practical applications. People with speech problems around the world have embraced them due to their intuitive design. Around 1% of Indians are in this group, which is a quite high percentage. It is for this reason that the incorporation of a framework familiar with Indian Sign Language would have such a profoundly positive effect on the lives of the people of India. A median filter is used to an input image to remove unnecessary details and improve clarity. Feature extraction is performed using principal component analysis (PCA), and the YCbCr color space is used for hand segmentation. The model is then trained through Regularized Extreme Learning. Using regularization to achieve peak structural performance for precise prediction, RELMs are a subclass of ELMs. This method exceeds popular alternatives like the support vector machine (SVM), Extreme Learning Machine (ELM), and CNN in terms of accuracy (around 97.8%).