Visual salient objects detection is an important fundamental application research of visual attention mechanism. It plays an important role in image retrieval

scene analysis

image annotation and object recognition. This paper proposes a novel approach for visual salient objects detection in natural scenes based on Treisman’s feature integration theory and Koch’s framework. In this approach

a visual saliency model for colored natural scenes is proposed and different feature saliencies are considered and computed. Then an effective method is given to extract salient objects from an integrated saliency map which is combined by different feature saliency maps. Comparing with Itti’s model

the experimental results indicate that not only the detection speed of this approach is faster

but also this approach can separate visual salient objects from their backgrounds more accurately and more efficiently. From this aspect

the approach in this paper is more similar to human’s real visual attention process than Itti’s model.