To implement real-time 3D reconstruction and displaying for polarization-modulated 3D imaging lidar system, an efficient subpixel registration based on maximum principal component analysis (MPCA) is proposed in this paper. With which only the maximum principal component is estimated to identify non-integer translations in spatial domain while other principal components affected by noise are ignored. Consequently, robustness and stability of the subpixel registration is implemented in presence of noise, while computational complexity is reduced and memory size is saved simultaneously, especially when the image size is large. Both simulated and real polarization-modulated images are used to verify the proposed method. Simulation results show that 0.01 pixels of the registration accuracy are implemented; meanwhile, experimental results show that the proposed method can effectively reconstruct the depth image in real world application.