Lip event detection using oriented histograms of regional optical flow and low rank affinity pursuit

摘要: An efficient oriented histograms of regional optical flow (OH-ROF) is presented to discriminatively code the visual appearance each lip motion frame.Each clip represented by a sequence OH-ROF vectors as its signature.We introduce stabilization scheme reduce impact irrelevant motions.We address an approach detecting silence event via small magnitude.We propose low rank affinity pursuit method determine lip-dynamic states mouth opening and closing. Lip detection crucial importance better understanding speech perceptually between humans computers. In this paper, we using pursuit. First, align extracted region sequences caused moving cameras. Then, field calculated from these sequentially stabilized images descriptor, namely OH-ROF, frame, whereby can be signature. Subsequently, detect based on magnitude, further that incorporates As result, various kind events appropriately estimated. The proposed neither requires any training set labeled videos nor learns priors in unconstrained video. Experiments show promising result comparison with state-of-the-art counterparts.