Abstract: Fractures are widely present in various kinds of rocks and are critical for reservoir evaluation, deve-lopment, and geological disaster prediction. Existing image-based fracture extraction technologies have noticeable deficiencies in micro-fracture recognition, fracture detail extraction, and fracture and pore differentiation. A multiscale fracture filter kernel superimposition noise reduction method was proposed, which used multiple filter kernels in superimposition to filter and reduce noises in CT scanning images of rock cores, with the filtering results then superimposed. A multiscale information-enhanced fracture segmentation method was also proposed. It combined image segmentation methods such as threshold and top-hat segmentation with edge enhancement to better capture micro-fracture information, applying different strategies for fractures of varying scales. The relationships between morphological parameters such as length, width, surface area, volume, shape factor, flatness, and elongation of the segmented 3D volumes were analyzed. The results showed that the proposed noise reduction method significantly reduced noise interference in CT scanning images of rock cores, and accurately preserved detailed fracture characteristics. This method improves the accuracy and stability of multiscale fracture extraction. The fracture morphological parameter chart effectively eliminates non-fractures such as pores and special mineral boundaries that are similar to fractures, achieving precise extraction of multiscale fractures.