It is difficult to capture a completely clear image of nonwoven web which is thicker than the depth of field of a light microscope. This phenomenon will leads to the data loss and the test error. In this paper, a region-based image fusion algorithm based on fibers natural boundaries was proposed. First, the one-pixel-wide boundaries were extracted from the point-based image fusion process. Then, the image fusion regions were formed by the sharpness diffusion from the source points which have the highest sharpness at local gradient within the boundaries of fibers. Finally, a fused clear image of the nonwoven web was constructed by replacing the regions with the corresponding regions which have the highest sharpness gradient from the series images at different focus positions of the light microscope.