Saliency Detection Based on Local and Global Information Fusion
Mengyue Ge, Ruirui Ji, Yi Wu · 2019
A saliency detection model based on local and global information fusion is proposed to improve the accuracy. Firstly, the saliency map reflecting local edge information is acquired based on the superpixel segmentation, saliency estimation and multi-scale linearly combination. Then the saliency map with better global information is obtained based on full convolutional network. Finally, two saliency maps are fused by the guided filtering and Hadamard product operation to improve the robustness of the saliency detection. The experimental results on MASR1000, ECSSD, and CDSS datasets show that the proposed fusion model could have higher Precision-Recall Curves and F-measure, and detect the salient object more accurately.