Abstract: According to the existence of local non-registered regions in the multi-focus images, a robust multi-focus image fusion algorithm based on guided filter and regional segment was proposed. First, a structure tensor based focus measure was imposed on the input images and the focus information was extracted. Second, an energy function was built to include the focus information of input images and was solved by graph cut to obtain the focus regions. Finally, the focus regions were filtered by the image guided filter to get the weight of each image and the fusion image was obtained in the gradient domain. Experiment results showed that the proposed method can obtain high quality multi-focus fusion image in the presence of mis-registration and is effective to preserve the edge of the image.