Gesture recognition based on improved shape context algorithm and Earth Mover's Distance

The common shape context algorithm does not have the rotational invariance property, which makes the characteristic extraction accuracy decreased a lot under some circumstances. A new improved shape context algorithm is proposed to solve this problem in this paper. The new algorithm chooses some key points and uses them as reference points, rather than like the traditional way that relies on all the contour information. Such improvement can make the new shape context algorithm rotation-invariant and can also simplify the original algorithm. Besides, another improvement is made in this paper. We combine shape context algorithm with EMD to create a new recognition method. The method is used for gesture recognition, and the experiment result shows that the new method has enhanced the gesture recognition accuracy a lot.