Indian Sign Language Recognition using Dynamic Region of Interest Extraction and Robust Features Extraction Algorithms

Abstract

The Indian Sign Language (ISL) recognition is complex tasks due to several reasons such as lack of sufficient ISL datasets, inefficient pre-processing and other image processing techniques to mitigate the challenges of illumination variations in raw images during the ISL recognitions, and poor accuracy. This research proposes novel hybrid solution depend on image processing terminologies self prepared scalable ISL dataset for all English alphabets and 0-9 digits. The proposed framework consists of keys steps such as pre-processing, Region of Interest (ROI) extraction, hybrid features extraction, and classification. After the pre-processing step, dynamic region growing method proposed to extract the accurate ROI from the input sign image. The accurate and fast segmentation leads to the overall system efficiency and robustness. The feature extraction is another important phase of image based recognition systems. The segmentation image is represented by unique set of features like visual features, texture features, gradient features etc. The limited set of features does not achieved the acceptable recognition accuracy, the hybrid features vector using the different image features with minimum computational efforts. For texture features we proposed more number of unique features to enhance the recognition accuracy further. The simulation results demonstrate the efficiency of proposed model on newly designed ISL dataset Comparison with state-of-art methods.

 Keywords: Features extraction, segmentation, recognition, Indian sign language, and pre-processing, visual features.