Image Fusion Algorithm Combining Saliency Detection And Structural Similarity

With the gradual development of imaging technology,the images obtained by any single sensor are not able to provide complete information on target scenes,and in some scenarios may be subject to some restrictions,such as infrared and visible light images.In order to better understand the situation of target scenes,the complementary information of these images must be concentrated into a single image.Image fusion technology is to combine useful information from multiple images of the same scene into one image,so that the fused images can take advantage of them and complement each other,provide more information about the scene than a single sensor,and reflect the characteristics of the target scene as completely as possible.Image fusion makes full use of the relevant information of the input images to enhance the efficiency of the image processing system and reduce the cost of the system.On the one hand,the image fusion based on structure similarity has good visual effects due to fully considering the structural information of images and the characteristics of the human visual system.However,this method is normally used to process the low frequency region of the image and to extract the structural information in the scene,and it has no good effect on the high frequency region of the image.On the other hand,the image fusion based on saliency detection can extract the information of high frequency such as edges and contours.By analyzing the advantages of the both of the image fusion methods,this paper develops a new image fusion algorithm which attempt to combine the similar structures and significant saliencies of source images.The experimental results of image fusion show the proposed algorithm have advantages over the similar algorithms available.The main contents and innovations of this paper are summarized as follows:1 In pixel-level image fusion,most image fusions are divided into three steps.Firstly,the image to be fused is decomposed into a low frequency region and a high frequency region,and then different fusion rules are used for image fusion in di:fferent regions.Finally,the low frequency region and the high frequency region are reconstructed into the final fused image.However,when performing image decomposition,The difference of the decomposition series will lead to different fusion results.Most of the multi-scale fusion methods are decomposed by more than two scales.These methods are computationally complex,so they require more memory and computation time.In this paper,a simple two-scale image decomposition is used to reduce the complexity of multi-scale fusion algorithm and to improve image fusion speed.2 After analyzing the advantages and disadvantages of structural similarity indicators and significance detection,then a new image fusion method by combination of saliency detection and structural similarity is proposed to obtain a better fused image.The method firstly decomposes the image into two-scale image,and adopts different fusion rules for the decomposed high-frequency region and low-frequency region respectively.A structurally similar image fusion is used for the low frequency region,and a significant detection fusion algorithm is used for the high frequency region containing a large amount of detailed information.Finally,the feasibility of the algorithm is verified by a large number of experiments.