Recognition of Single Handed Sign Language Gestures using Contour Tracing Descriptor

— Recently, many researchers have focused on building dual handed static gesture recognition systems. Single handed static gestures, however, pose more recognition complexity due to the high degree of shape ambiguities. This paper presents a gesture recognition setup capable of recognizing and emphasizing the most ambiguous static single handed gestures. Performance of the proposed scheme is tested on the alphabets of American Sign Language (ASL). Segmentation of hand contours from image background is carried out using two different strategies; skin color as detection cue with RGB and YCbCr color spaces, and thresholding of gray level intensities. A novel, rotation and size invariant, contour tracing descriptor is used to describe gesture contours generated by each segmentation technique. Performances of k-Nearest Neighbor (k-NN) and multiclass Support Vector Machine (SVM) classification techniques are evaluated to classify a particular gesture. Gray level segmented contour traces classified by multiclass SVM achieve accuracy up to 80.8% on the most ambiguous gestures of ASL alphabets with overall accuracy of 90.1%.