Person authentication by air-writing using 3D sensor and time order stroke context

This paper proposes touch-less system to use air-written signatures for person authentication. This smart system can verify a person’s ID without using any mouse, touch panels, or keyboards. It benefits from a 3D sensor to capture a user’s signature in air and then verifies his ID via a novel reverse time-ordered shape context. This backward representation can effectively filter out redundant lifting-up strokes and thus simplify the matching process as a path finding problem. As features to analyze a user’s signature more accurately, the rates of turning point and curvature are also embedded into this representation. Then, with a weighting scheme, the path finding problem can be solved in real time via a dynamic time warping technique. Another challenging problem is the multiplicity problem which means a signature is not always written the same due to users’ practices and moods. Thus, an agglomerative hierarchical clustering scheme is adopted to cluster users’ signatures into different subclasses. Each subclass represents different within-class variations. Another key issues is the criterion to determine the threshold for verifying and then passing a user’s signature. Experimental results proves the average within-class distance can gain the best accuracy. The proposed solution achieves quite satisfactory authentication accuracy (more than 93.5%) even though no starting gesture is required.