Abstract: In order to improve the identification ability of targets, the fusion of infrared and visible images is widely applied to the night vision system. With Compressive Sensing (CS), the complete information of signal can be kept by small amounts of linear projection, such that the contradiction between the cost of infrared detector and the image resolution can be resolved. The reconstruction error and calculated amount can be reduced, if the image fusion can be performed based on the measurements in CS directly. Thus, this paper proposes a novel fusion technology of infrared and visible images in CS. Firstly, our algorithm proposes a new cost function, which considers both conformity degree between the fused image and original images and the preservation degree of original image features. Secondly, the sparse coefficients of fused image can be acquired by L₁-norm optimization. Finally, the fused image can be reconstructed with the sparse coefficients and the dictionary. In comparison with several fusion methods in CS, our algorithm has a significant advantage in terms of several metrics, as well as in the visual quality. This method provides a new useful tool for the image fusion in CS.