Most existing saliency detection methods utilize low- level features to detect salient objects. In this paper, we first verify that the foreground objects in the scene can be an effective cue for saliency detection. We then propose a novel saliency detection algorithm which combines low level features with high level object detection results to enhance the performance. For extracting the foreground objects in a scene, we first make use of a camera array to obtain a set of images of the scene from different viewing angles. Based on the array images, we identify the feature points of the objects so as to generate the foreground and background feature point cues. Together with a new K-Nearest Neighbor model, a cost function is developed to allow a reliable and automatic segmentation of the foreground objects. The outliers in the segmentation are further removed by a low-rank decomposition method. Finally, the detected objects are fused with the low-level object features to generate the saliency map. Experimental results show that the proposed algorithm consistently gives a better performance compared to the traditional methods.