Application Of Visual Attention Mechanism To Remove The Background

Visual attention is a important path for human to obtain information. When the things stimulate our eyes, human vision relies on visual attention mechanism to select or keep the useful parts of scene, on which the complex information can be effectively or reliably processed. With development of visual attention, the model of visual attention not only testify the theory of human visual system, but also help us select problem and enhance the fast of the computer run. Now, it also has a useful value in the field of image understanding and analysis, object detection, image compress. But natural scenes are usually composed of several dynamic entities. Foreground objects often move amid complicated backgrounds that are themselves moving, such as swaying trees or other objects such as a crowd, a flock of birds, moving water, waves, snow, rain and smoke-filled environments. Even for static scenes, egomotion of image sensor can cause a highly variable background. In most extreme situation, egomotion and scene motion can combine to produce very complex background motion patterns. We refer to scene with any of these types of variability as dynamic scene. Since such scenes are plentiful in the natural world, successful discrimination between the background motions they induce and moving foreground objects will become very important, but the current algorithm of background subtraction is not applicable to scenes with dynamic backgrounds. Therefore, the background subtraction algorithm based on visual attention is proposed in this paper.The proposed algorithm mostly based on the ideal and method of visual attention computing. Background subtraction is equated to the dual problem of saliency detection, the current frame is first processed before background subtraction and compute its saliency, background points are those considered not salient by suitable locally contrast. Under this formulation, the saliency of a location is determined by the locally contrast which opposes center to surround, the features of locally contrast are modeled as dynamic textures, the combination of discriminant center-surround saliency with the modeling power of dynamic textures yields a robust, fast and fully unsupervised background subtraction algorithm, applicable to scenes with highly dynamic backgrounds.The paper designs two different experiments based on ideal of visual attention, with static background and dynamic background scene, respectively. The experiment results show that well overcome the drawback of previous methods. Such as, static cameras, compensation of camera motion, foreground objects that move in a consistent direction or have faster variations in appearance than the background, explicit background models. In this experiments, effectively detecting the moving targets and removing the background, the average error rate has also been significantly reduced. The mechanism of visual attention is more likely to the human visual system.