In order to improve the visual effect of infrared and visible light image fusion and overcome the artifact effect of the fusion result, an image fusion method based on the internal generative mechanism and convolution sparse representation is proposed. Firstly, the source image is decomposed using the internal generative mechanism that conforms to the reasoning of the human brain to obtain the prediction layer and the detail layer. Then, the detail layer is decomposed using a convolution sparse representation to obtain the secondary detail layer and the basic layer, and the activity level measurement is made to be larger and the weighted average rule is fused separately. The ISR hybrid operator fusion rule is defined for the prediction layer. Finally, the fusion prediction layer and detail layer are added to obtain the final fusion result. In the experiment, three representative infrared and visible light images are used for algorithm testing. The experiment results show that the proposed method has good subjective visual effects, and the objective evaluation indicators are also better and effective.