A Robust Method for Microscopic 3D Shape Restoration via Shape-From-Focus.

Shape-from-focus (SFF) is essential for microscopic three-dimensional morphology analysis due to its non-contact nature and high axial resolution. However, in practical microscopy environments, ambient vibration and multi-source noise often cause unstable sampling and distorted focal volume, significantly limiting the technique's applicability in high-precision measurement. To address this, we propose a novel anti-vibration, high-fidelity microscopic SFF framework. First, a temporal redundancy strategy based on video streaming is employed to generate spatially consistent image sequences prior to reconstruction. Subsequently, an adaptive dual-mode peak localization algorithm is introduced, which intelligently selects depth estimation strategies according to the texture characteristics of the micro-surface, achieving sub-pixel accuracy while maintaining computational efficiency. Furthermore, by transforming one-dimensional focus signals into two-dimensional images and applying an intelligent classifier for filtering, the method successfully separates noise from valid texture features. This framework requires no prior knowledge of vibration frequency and adaptively suppresses random environmental disturbances. Experiments on both synthetic and real microscopic samples show that the proposed method outperforms existing approaches in shape recovery accuracy, preservation of fine texture details, and noise suppression. The technique offers a reliable high-precision 3D measurement solution for materials science, industrial inspection, and biomedical microscopy in vibration-prone environments.