Interferometric synthetic aperture radar (InSAR) is the preferred technique for monitoring three-dimensional (3D) surface displacements. However, conventional InSAR phase unwrapping (PhU) methods face challenges in interpreting local low-coherence regions and dense fringe patterns, reducing the reliability of subsequent 3D decomposition. Additionally, the North-South (N-S) displacement retrieved from InSAR measurements is often inaccurate due to the near-polar orbits of radar satellites. To address these issues, this paper presents a novel method for retrieving mining-induced 3D displacements from dual InSAR geometries. First, we propose a two-step PhU approach based on the central difference information filter phase unwrapping (CDIFPU), which can generate a complete deformation field. Subsequently, vertical and East-West (E-W) displacements are separated using InSAR measurements from dual imaging geometries. Finally, we use an extensible subsidence facet kernel to retrieve the N-S displacement based on random medium theory. Simulated experiments show that the two-step PhU improves unwrapping accuracy by approximately 78.4%, 64.8%, and 59.5% over the branch-cut, MCF, and CDIFPU methods. The accuracy of the two-step PhU in retrieving vertical and east-west displacements is 65.8% and 71.3% higher than that of the MCF. Moreover, the subsidence facet kernel further improves the N-S displacement accuracy by 68.5% compared to the conventional first-order difference operators. Experiments with RADARSAT-2 and TerraSAR-X images from the Fengfeng Mining Area, China, demonstrate that the two-step PhU is more effective than SNAPHU, PUMA, and one-step deep learning PhU in retrieving absolute phase in local low-coherence regions, enhancing the quality of vertical and E-W displacements. Ultimately, its N-S displacement is successfully extracted using the subsidence facet kernel.