Environment enhanced fusion of infrared and visible images based on saliency assignment

Among existing region-based infrared (IR) and visible (VIS) fusion methods, source images are segmented into thermal targets and backgrounds. Background areas typically have a small grayscale range and are less contrasty, which makes nontarget objects difficult to discern, reducing visual effects of the fused image. To solve this problem, an environment enhancement method based on saliency measure for IR-VIS fusion is proposed. This paper analyses the relationship between gray level of IR data and semantic importance information. The Fuzzy C-means algorithm is used to divide IR images into different ranks of semantic importance, which can provide the most accurate semantic representation of an IR image. Moreover, to maintain the thermal targets highlighted, an effective enhancement strategy termed saliency assignment, is proposed so that the low-level features are arranged to provide viewers with directed attention. Finally, a weighted averaging fusion algorithm based on the importance ranks, to obtain fused images. The experiment is performed on common datasets, which consists of subjective and objective tests, and proves the validity of the proposed algorithm.

Automated summary

In this paper, a saliency-based method for infrared and visible light fusion is proposed. By segmenting the images into thermal targets and backgrounds and enhancing them based on semantic importance information, the visual effects of the fused image are improved.