The invention discloses a robustness image fusion method based on a wavelet coefficient statistics model. The robustness image fusion method of the invention comprises steps of preparing two noisy images to be fused, respectively performing dual-tree complex wavelet transformation on the two noisy images, decomposing the two noisy images into low frequency sub-bands and high frequency sub-bands, extracting contraction operators from the high frequency sub-bands of the two images, establishing a combined probability distribution and an independent edge distribution between two coefficients to be fused, establishing a matching degree based on the combined distribution and the edge distribution, proposing significance measurement on the basis of edge distribution and on the basis of divergence and information entropy and combining with the contraction operator. Under a statistic model frame, the robustness image fusion method puts forward a parameter estimation method using the moment estimation, replaces the moment in the fusion algorithm with de-noising parameters in a de-noising algorithm, and enables the fusion rule frame not to be affected by noise. Meanwhile, the robustness image fusion method combines a de-noising operator with a fusion weight, maintains detail information of a source image while accurately removing noise, and further improves a de-noising effect of the fusion algorithm.