Robust Foreground Extraction Technique Using Gaussian Family Model and Multiple Thresholds Hansung Kim 1 , Ryuuki Sakamoto 1 , Itaru Kitahara 1, 2 , Tomoji Toriyama 1 , and Kiyoshi Kogure 1 1 Knowledge Science Lab, ATR, Kyoto, Japan 2 Dept. of Intelligent Interaction Technologies, Univ. of Tsukuba, Japan {hskim,skmt,toriyama,kogure}@atr.jp, kitahara@iit.tsukuba.ac.jp Abstract. We propose a robust method to extract silhouettes of fore- ground objects from color video sequences. To cope with various changes in the background, the background is modeled as generalized Gaussian Family of distributions and updated by the selective running average and static pixel observation. All pixels in the input video image are classi- fied into four initial regions using background subtraction with multiple thresholds, after which shadow regions are eliminated using color compo- nents. The final foreground silhouette is extracted by refining the initial region using morphological processes. We have verified that the proposed algorithm works very well in various background and foreground situa- tions through experiments. Keywords: Foreground segmentation, Silhouette extraction, Back- ground subtraction, Generalized Gaussian Family model. 1 Introduction The background subtraction technique is one of the most common approaches for extracting foreground objects from video sequences [1,2]. This technique sub- tracts the current image from a static background image acquired in advance from multiple images over a period of time. Since this technique works very quickly and distinguishes semantic object regions from static backgrounds, it has been used for years in many vision systems such as video surveillance, tele- conferencing, video editing, and human-computer interfaces. Conventional approaches assume that the background is static; therefore, they cannot adapt to changes in illumination or geometry in it [3,4,5]. Several algo- rithms have been developed to overcome this problem by modeling and updating the background statistics. They can be classified into two categories: parametric and non-parametric approaches. The parametric approaches set a form of the background distribution in ad- vance and estimate the parameters of the model. Earlier methods used single Gaussian distribution to model the probability distribution of the pixel in- tensity [6,7]. Recently, the Gaussian mixture model is the most representative Y. Yagi et al. (Eds.): ACCV 2007, Part I, LNCS 4843, pp. 758–768, 2007. c  Springer-Verlag Berlin Heidelberg 2007