Abstract: For solving the problem that the current fusion rules lack uniformity for the approximation subband and the detail subbands in the multiscale image transform domain, a novel approach to image fusion is proposed based on statistical mixture modeling in pyramidal dual-tree directional filter bank(PDTDFB) transform domain. Firstly, coefficients of the approximation subband, the high frequency subband and all bandpass directional subbands, are respectively modeled by Gaussian mixture distribution. Then, their distribution parameters are estimated by the expectation maximization algorithm, and all subbands are fused by the new rule using the weighted average of a posteriori probability. Finally, an effective fused image is achieved through inverse PDTDFB transform. The experimental results demonstrate that, comparing with existing multiscale fusion approaches, the proposed method uses uniform fusion rule for all subbands, makes the fused image have better performances, and can be fit for all kinds of image fusions.