Visual saliency maps of the source images are extracted with various image features.A novel fusion rule for the low-frequency subbands of multi-scale image fusion algorithms is proposed,and a new multi-scale image fusion method is developed based on the visual saliency maps.trous wavelet is integrated with Nonsubsampled Contourlet Transform,and is applied to the experiment of multisensor and multi-focus image fusion.It is demonstrated that the proposed approach yields better results both in visual inspection and objective evaluation than the methods choosing low frequency coefficients based on average or neural network.

With the knowledge that the locally maximum of the dim target gray value usually shows a sudden change of the spectrum in frequency domain,a two-stage infrared dim target enhancement approach based on spectral analysis and image fusion is proposed.The algorithm extractes the spectral residual according to the spectral difference between the dim target and the image background,constructed the saliency map of the potential targets,and applies different fusion rules to potential targets and background respectively with the information redundancy of continuous frames to obtain the fused image.Experimental results showe that the method can enhance the size and the total energy of the infrared dim target effectively and efficiently.

Multi-sensor image fusion has attracted much attention recently. Although it is relatively simple to obtain a fused image, evaluating the performance of fusion algorithms is much harder in practice. This paper presents a novel non-reference objective quality metric for night vision image fusion. The metric takes human perception and the imaging characteristics of the infrared and visible images into account, divides the source images into different local regions, and the quality of fused image is evaluated by regional information similarity, instead of computed pixel by pixel as traditional methods. Experimental results show that this metric is consistent with human visual inspection and can be applied to compare image fusion schemes that are not performed at the same level.

Based on the region partition and association of source images,two regional features of source images are defined according to the imaging characteristics of infrared and visible images and the requirements of fusion tasks.A regional similarity assessment index with the regional features and the information entropy and mutual information of fused images and source images is constructed.A novel non-reference quality evaluation metric for infrared and visible image fusion is proposed.Experimental results show that this metric fits the results of human visual inspection better than the recent state of the art image fusion evaluation metrics.