作者: Mohammad Sadeghi , Josef Kittler , Kieron Messer
关键词: Mixture model 、 Artificial intelligence 、 Scale-space segmentation 、 Computer science 、 Computer vision 、 Segmentation 、 Pixel 、 Cluster analysis 、 Initialization 、 Image processing
摘要: We propose a novel segmentation method for real time lip tracker initialisation which is based on Gaussian mixture model of the pixel data. The built using Predictive Validation technique advocated in [4]. In order to construct an accurate time, we adopt quasi-random image sampling Sobol sequences. test proposed database 145 images and demonstrate that its accuracy, even with few number samples, satisfactory significantly better than obtained by k-means clustering. Moreover, does not require segments be specified priori.