The structure of nonwoven fabric has a great influence on the fabric properties, and the filtration performance of the fabric is determined by its pore structures and key parameters such as pore size, distribution and porosity, which should be described accurately and appropriately. However, at present problems exist in the methods of describing the three-dimensional structure of nonwovens, such as the high cost, slow speed and inaccurate model. This paper proposes a new method of reconstruction algorithm, inspired by the criss-cross internal structure of nonwovens. This nonwoven structure with a certain depth presents some implication of the processing information, which is similar to the formation of an anthill. The reconstructing processes are as follows. First, the multi-focused images of fibers are captured by the microscope and the images are then fused into a two-dimensional image. Second, the three-dimensional coordinate of the source point is calculated by the fiber segmentation and One-to-Many projection. Finally, the fiber axis is generated by connecting the source point according to the prescribed conditions, and then the closed three-dimensional surface is the three-dimensional path of a dynamic-radius sphere rolling along the fiber axis. The experiment results demonstrate that the proposed method can reconstruct the accurate structure of the nonwoven fabric and retain its vertical connection information, which is of great significance to the interception of harmful substances, the production and preparation of filter medium and the follow-up study of the filtration mechanism.