Weighted feature correlation and fusion saliency

Abstract

In this paper, we propose a novel feature correlation and fusion approach for multiple visual focuses content associational problem. Integrating various visual attention models to extract the visual focus of the image in the visual database, a weighted fusion of visual focuses will be obtained in good accuracy and the corresponding visual focus set will also be built subsequently. Then, the correlation matrix based on normalized mutual information and structural similarity index measurement will be computed within visual focus set. Through scanning correlation matrix in turn, the corresponding focus fusion process will be carried on and we use the weighted saliency model to compute visual focus of fusion focus. Compared with the state-of-the-art methods such as Itti, IS, GBVS, IF, NCS, MS and ISRW, higher robustness and accuracy rate are the main outstanding advantages of the presented approach. Experimental results on high noise interference confirm the validity of our approach.