Template selection based superpixel earth mover's distance algorithm for hand gesture recognition

Abstract

In this paper, we propose a new template selection based superpixel earth mover's distance (TS-SP-EMD) algorithm for hand gesture recognition. In the original SP-EMD, template matching is utilized as the classification method. Therefore, the quality and quantity of templates are closely related to the recognition accuracy and computational speed. To address this issue, we propose a A-medoids based template selection method to choose the key gesture samples, i.e. the templates, from a group of candidates. With the proposed method, the number of templates are reduced thus the recognition speed can be increased with a small cost of recognition accuracy. Experimental results using public gesture datasets show that the proposed template selection method can largely reduce the number of required templates, with only slightly drop of the recognition performance.