Abstract:The existing image fusion methods have edge ladder effect in varying degrees, which will lead to some spatial artifacts introduced into the fused image. This paper proposes a new method to solve the poor robustness in the process of image fusion, that is, forward-backward Self-correcting diffusion guided feature reconstruction (Forward-backward Self-correcting diffusion, FBSD). The algorithm fully considers the gradient variance of the transition between features, and suppresses the ”heat transfer” of pixels in all directions according to the threshold of diffusion coefficient. In order to solve the problem of edge sharpening caused by backward diffusion, we introduce the decomposition method of variable index to control the diffusion direction and limit the backward diffusion to a limited range, so as to effectively prevent the formation of ladder effect. According to the differences between features after decomposition, a hybrid fusion strategy based on Expectation-Maximization algorithm and principal component analysis algorithm is designed. Finally, the evaluation index is used to evaluate the performance of the proposed algorithm, and it is verified that this method is better than the existing image fusion methods in dealing with the edge step effect, and the effectiveness of the fusion decision.