Dual Searching Window Based Face Tracking

Abstract

In this paper, an enhanced CAMShift (continuously adaptive mean shift) object-tracking algorithm, called dual searching window based face-tracking algorithm, is proposed. Firstly, the near complexion background is removed from video frames by using a mixture Gaussian model. An accessory window is then introduced to correct the tracking error resulting from temporary occlusion by an object with similar probability distribution. Experimental results have testified the proposed face-tracking scheme can greatly alleviate the flesh-like background interference and miss tracking due to the temporary occlusion