Hand Gesture Recognition Using Interactive Image Segmentation Method

Abstract

In this paper, a novel hand gesture recognition method using interactive image segmentation algorithm is proposed. We applied Gaussian mixture model to build the model of color image and the iteration of expectation maximum algorithm learnt the parameters. Then the graph model of color image is built. Finally, the segmentation is achieved by minimizing the energy of graph model according to min-cut/max-flow algorithm. Segmentation results were quantitatively tested and compared, by evaluate the region accuracy and boundary accuracy of segmentation results. To apply interactive image segmentation method into a fully automatic recognition framework, we applied human skin feature and depth information to generate the initial seeds. We also built a hand gesture database which contains ten kind of hand gestures for recognition test, proving that the segmentation of hand gesture images improved the recognition accuracy.