Abstract: To overcome the drawback of the semi-supervised segmentation methods,we propose a new saliency based automatic natural object segmentation method.We choose the spectral residual method as the saliency method for its low cost.We make erosion and dilation on the coarse segmentation region to get the object label and background label automatically.By maximal similarity based region merging with the help of labels,the object can be effectively extracted from the background.Extensive experiments results on a large variety of natural images confirm that our framework can reliably and automatically extract the object from the complex background.