Abstract: An infrared feature and image similarity-based image fusion scheme, which combines aspects of feature-level fusion with pixel-level fusion, is proposed to improve the computing speed and the quality of the fused image, thus making it easier for the observer to interpret the scene. In the proposed scheme, the infrared image is segmented based on the infrared features of the targets, and the target regions and the background are fused respectively; in the fusion process of the target regions, the image structural similarity is used to measure the similarity between the corresponding target regions in the source images, on the basis of which different fusion rules are applied to the redundant and complementary/conflicting regions. Compared with some commonly used multiresolution image fusion methods, the scheme presented is much faster, with an improved fusion effect, which is due to the full use of the infrared features of the infrared image and the detailed information of the visible image.

Key words: image fusion, infrared feature, image structure similarity, region similarity