HIERARCHICAL FOREGROUND SEGMENTATION BASED ON A SPATIO-TEMPORAL GAUSSIAN MIXTURE MODEL IN TRANSFORM DOMAIN
Yuki Hishinuma, Hiroaki Tezuka, Takao Nishitani · ITC-CSCC :International Technical Conference on Circuits Systems, Computers and Communications · 2009
This paper describes a hierarchical unified parameter generation on fine-to-coarse multi-resolution foreground segmentation based on Gaussian mixture model (GMM). The hierarchical parameter generation can reduce the required update processing amount by half in the coarse processing area, without serious performance degradation on high resolution video pictures. The GMM approach over Walsh transform domain in every resolution processing is modified to a set of finer resolution layers by the GMM approaches, followed by the remaining set of coarse resolution layers using single Gaussian models which employ the unified parameters generation scheme. The unified parameter generation is based on the reproducing property of Gaussian distributions used for GMM parameters in the last GMM layer. Experimental results show almost the same stable performance of the fine-to-coarse GMM in outdoor conditions.