Abstract: The coseismic 3D surface deformation field is critical data for quantifying tectonic activity, inverting focal mechanisms, and assisting post-earthquake emergency response, holding significant value for earthquake research and disaster prevention and mitigation. Interferometric Synthetic Aperture Radar (InSAR) technology provides only one-dimensional (1D) deformation measurements along the satellite line-of-sight (LOS) direction, whereas recovering coseismic 3D surface deformation using SAR data requires at least three independent viewing geometries to construct the observation equations. Existing 3D coseismic deformation inversion algorithms commonly suffer from insufficient accuracy in near-fault zones and the loss of fine deformation details when dealing with surface-rupturing earthquakes. To address the limitations of the existing weighted least squares (WLS) algorithm and the stress-strain model with variance component estimation (SM-VCE) algorithm, this paper proposes a joint estimation method for coseismic 3D surface deformation fields considering pixel encoding classification. Based on deformation characteristics, the proposed method implements homogeneous-point encoding and partitioning across the entire observation field to suppress the cross-fault mixing of heterogeneous data during the windowing process of the conventional SM-VCE algorithm, thereby better preserving the step-like deformation details and surface rupture morphology in the rupture zone. Synthetic experiments and a case study of the 2021 Ms7.4 Maduo earthquake in Qinghai demonstrate that the proposed method reduces the root-mean-square error (RMSE) of the inverted horizontal deformation by 35.5% compared to the traditional SM-VCE method. In the real-world application to the Maduo earthquake, the RMSE between the inverted horizontal deformation and GNSS measurements reaches 6.82 cm, outperforming the SM-VCE algorithm. The proposed method effectively recovers near-fault step-wise deformation caused by surface-rupturing earthquakes, providing a more complete and reliable coseismic 3D deformation field.