Robust Pose Invariant Shape-based Hand Recognition

This paper presents a novel technique for hand shape and appearance based personal identification and verification. It has two major building blocks. A segmentation block presents robust and fully automatic algorithms which are able to accurately segment the hand's palm and fingers irrespective of colour contrast between the fosreground and background. They achieve a consistent representation of the fingers and the palm regardless of their pose/orientation or the spaces between the fingers. In the feature extraction/matching block, the iterative closest point (ICP) algorithm is employed to align the images. Both shape and appearance based features are extracted and comparatively assessed. The modified Hausdorff distance and independent component analysis (ICA) algorithms are used for shape and appearance analysis. Identification and verification were performed using fusion strategies upon the similarity scores of the fingers and the palm. Experimental results show the proposed system exhibits an accuracy of over 98% in hand recognition and verification in a database consisting of 500 different subjects.