Abstract: To overcome the shortcomings of the traditional popular graph-based sorting algorithms，only using the
image boundary as the background query，the accuracy of the query selection directly affects the results of the
algorithm，an improved algorithm is proposed． Based on the detection results of current algorithms，the selection
of foreground and background seeds is optimized． Firstly，super-pixel segmentation is performed on the image to
make full use of the middle-level information of the image． Secondly，the image saliency map is calculated with
the popular ranking algorithm． Finally，the saliency result is processed to select better query points and obtain
the final saliency map． Compared to eight algorithms on CSSD( Complex Scene Saliency Datase) and ECSSD
( Extended Complex Scene Saliency Datese) datasets，the experimental results show that the proposed algorithm
has higher detection accuracy．

Key words: manifold ranking, salient object detection, query optimization