Saliency Technique for Efficient Back Ground Subtraction

Aiswarya Muralidharan, S. Sivakumar

Abstract: The key technique for automatic video analysis is Background subtraction, especially in the domain of video surveillance. In this paper Saliency detection can be used in background subtraction. The proposed method is effectively used in the case where the scene is highly changing or the camera is not moving. The separation of foreground and back ground regions within and across video frames is done by visual and motion saliency information. Input video contains all this information which can be extracted. The saliency induced features is effectively combined by conditional random field (CRF) which deals with unknown pose and scale variations of the foreground object. The method is computationally efficient, reliable, and simple to implement and thus it can be easily extended to various applications. It is effective in detecting saliency compared to various State-of-the-art methods.

Keywords: Saliency detection, change detection, Condition random field, background subtraction,