Color Invariant Study for Background Subtraction

Effectiveness detection to extract objects of interest is a fundamental step in many computer vision systems. In real solutions, the accurate Background Subtraction (BS) is a challenge due to diverse and complex background types. Being the color widely used as descriptor to improve accuracy in several BS algorithms, in this paper we analyze four Color Invariants (CIs) based on the Kubelka-Munk theory combined with Gray scale. The capability of several CIs combinations in segmenting foreground is evaluated referring to five video sequences. This experimental study provides a point-of-view to choose the best color combination considering accuracy and the channel numbers which can be applied for image segmentation. The results demonstrate that the combination of the color invariant H with Gray scale achieves higher performance for foreground segmentation for both indoor and outdoor video sequences. Furthermore, it uses the minimum number of color channels.