Hair-color model and adaptive contour templates based head detection
Min Zhao, Dihua Sun, Wan-mei Fan · 2010
A novel method for head detection is proposed dealing with video sequences captured with fixed mono- camera, which constructs two detectors utilizing the hair-color and the contour features respectively. This algorithm implements head detection correctly combining hair-color and head contour features together rather than independently applying color detector and contour detector respectively. Firstly, hair-color probability density was modeled, which directs image segmentation for the purpose of obtaining candidate object region and abstracting corresponding features. Secondly, with the features of the candidate objects, the contour templates of each candidate were shaped automatically, which confirm the head target finally when templates match. Experimental results indicate that the algorithm presented resists false detection of objects whose color distribution is similar to hair color, and therefore improve the accuracy.