To preserve image edge and background information during the fusion of infrared and visible images, an improved co-occurrence filtering algorithm is proposed in this paper. It is first analyzed that the co-occurrence filtering effect depends on the filter scale standard deviation and the image content. Secondly, the filtering scale is optimized based on skewness of pixel intensity and pixel energy, and filtering speed is improved by adjusting the threshold for pixel-to-pixel distance. Finally, the NSST algorithm uses different fusion methods for the low and high frequency subbands of the image, and the low-frequency subbands of the image are fused using Delaunay interpolation calculation, the maximum symmetric surround saliency method, and the sum modified Laplacian method is used for fusion of high-frequency sub-band images. The experimental simulation demonstrates that the fusion results of this paper's algorithm are richer in the edge detail part, no artifacts appear, and have obvious advantages in both subjective vision and objective indexes.