Because the image sensor has different interference in the process of acquisition and quantification transmission, for example, the low-light image will be interfered by the inconsistent gain of the microchannel plate and the dark current effect. In the process of infrared image acquisition, it is affected by atmospheric transmission, the uniformity of detector arrays, external randomness and human interference. These factors make the acquired images often contain more or less sound. In order to avoid image noise degrading the quality and performance of subsequent image fusion, the source image must be denoised before image fusion. In addition, due to the different kinds of detectors, the low-light and infrared source images must be effectively and accurately registered before pixel-level image fusion to remove the systematic and accidental errors of different sensors. The registration accuracy between each source image will directly affect the image fusion effect. Pixel-level image fusion requires the registration accuracy of one pixel between source images.