Delayed absolute difference (DAD) signatures of dynamic features for sign language segmentation

In sign language segmentation, individual gestures are extracted out of a continuous stream and then matched with models for recognition. We have hypothesized an improvement in word segmentation without affecting the language naturalness by incorporating a novel set of segmentation features (pause, repetition and directional variations). To analyze these segmentation features, a unified tool (DAD signature) is presented that encodes the segmentation features in form of distinct patterns. It is shown that the DAD signature can easily detect the pauses, repetitions and reversal of direction.