A fracture surface guided multistage registration framework for bamboo slip point clouds

Fragmented bamboo and wooden slips are difficult to reassemble digitally because their ultra-thin geometry, narrow fracture surfaces, abrasion, and material loss provide weak and unreliable geometric constraints. We present JianduReg, a fracture-surface-aware coarse-to-fine registration framework for virtual reassembly. Fracture regions are first obtained through semi-automatic cropping and expert-guided correction, restricting registration to physically meaningful broken interfaces. A normal-enhanced Deep Closest Point module jointly encodes coordinates and surface normals to estimate a stable coarse pose. A bidirectional maximum correntropy criterion ICP module then refines alignment by down-weighting unreliable correspondences, noise, and missing regions. Experiments on ModelNet40, synthetic bamboo-slip fragments, and real scanned Jiandu artifacts show improved registration stability, tighter fracture-surface closure, and reduced tangential slipping compared with traditional, learning-based, and robust-kernel baselines. JianduReg therefore supports non-contact digital restoration of fragile written heritage.