Abstract: Aiming at the problem that traditional saliency detection methods suffer from unclear boundary and bad robust detection performance,an novel image salient regions detection method was proposed,combining two detection stages including edge boxes for rough location and low-rank background model for refining,to enhance the performance of salient regions detection.First,it improves the image salient regions detection method based on edge boxes.It uses OTSU method for adaptively computing the optimal threshold value of edge magnitude,to replace fixed threshold method and reduce boundary detection error.Second,on the basis of suspicious salient regions detected by edge boxes based method,it uses robust principal component analysis method to obtain the low-rank component of the image for building a background model,and eliminates the background regions based on background subtraction method to reduce false detection of salient regions.Experimental results on the PASCAL VOC 2007 dataset show that,this method can significantly improve the precision and recall metrics of salient regions detection,and has higher detection efficiency.

Key words: Salient regions detection,Edge boxes,Robust principal component analysis,Low-rank background,OTSU