Comparative Analysis of Feature Detection and Extraction Techniques for Vision-based ISLR system

Sign language recognition is a highly adaptive interface between the deaf-mute community and machines. In India, Indian Sign Language (ISL) plays a significant role in the deaf-mute society, breaking communication distancing. Extracting features from the input image is crucial in vision-based Indian Sign Language Recognition (ISLR). This paper addresses feature detection and extraction techniques used in the ISLR. This paper categorizes existing techniques into three broad groups: scale-based, intensity-based, and hybrid techniques. SIFT (Scale Invariant Feature Transform), SURF (Speeded up Robust Features), FAST (Features from Accelerated Segment Test), BRIEF (Binary Robust Independent Elementary Features), and ORB (Oriented FAST and rotated BRIEF) are the techniques that have been evaluated and compared for intensity scaling, occlusion, orientation, affine transformation, blurring, and illumination. Results were generated in terms of key point detected, time-taken, and the match rate. SIFT is consistent in most circumstances, though it is slow. FAST is the fastest with good performance like ORB, and BRIEF shows its advantages in affine transformation and intensity changes.