@article{Akilan2018414,
  title = "Fusion-based foreground enhancement for background subtraction using multivariate multi-model Gaussian distribution ",
  journal = "Information Sciences ",
  volume = "430–431",
  number = "",
  pages = "414 - 431",
  year = "2018",
  note = "",
  issn = "0020-0255",
  doi = "https://doi.org/10.1016/j.ins.2017.11.062",
  url = "https://www.sciencedirect.com/science/article/pii/S0020025516320539",
  author = "Thangarajah Akilan and Q.M. Jonathan Wu and Yimin Yang",
  keywords = "Background-foreground detection",
  keywords = "Multivariate multi-model Gaussian distribution",
  keywords = "Color distortion",
  keywords = "Color similarity",
  keywords = "Illumination changes",
  keywords = "Feature fusion ",
  abstract = "Abstract These days, detection of visual attention regions (VAR) such as moving objects have become an essential pre-processing stage in many computer vision applications. In this paper, we focus on the vital issue of separating moving objects a.k.a. Foreground (FG) in a scene, which has a near-static background (BG). We address the difficulty in setting an adaptive threshold in the multi-model Gaussian-based BG-FG separation through a novel \{FG\} enhancement strategy by assimilating color and illumination measures. We formulate the problem mathematically by using a histogram of a fused feature of color and illumination measures. The proposed method improves the BG-FG separation by introducing the following items: (i) A new distance measure to check if a pixel matches a Gaussian distribution. (ii) A new strategy to enhance the results of traditional background subtraction (BGS) with a fusion of color and illumination measures. (iii) A methodology to find appropriate threshold adaptively that separates \{BG\} and FG. (iv) A foreground validation process through probability estimation of multivariate Gaussian model distribution (MVGMD). We test the proposed algorithm on five different benchmark video sequences. The experimental results reveal that the proposed approach works well in challenging conditions, at the same time, it performs competitively against state-of-the-art Gaussian-based algorithms and few other non-Gaussian-based methods as well. "
}