Abstract: To address the issue of low correspondence accuracy and poor bijection caused by ignoring the consistency of mapping direction in the calculation of the correspondences between existing non-rigid 3D models, a method to optimize the function mapping between the spectral and spatial domains is proposed. First, the need for manually designed markers is eliminated by the spectral method and spectral feature descriptors that are independent of the need for manually designed markers are computed, while spectral feature descriptors that are independent of the model’s symmetry and maintain consistent directions are computed as well. Secondly, the function mapping is improved by optimizing the correspondence between the spectral and spatial domain, resulting in high-quality mapping outcomes. Finally, an iterative optimization approach is used to align the function mapping matrix, leveraging the model’s inherent symmetry to calculate accurate correspondences between non-rigid 3D models. Experimental results demonstrate that the proposed algorithm achieves optimal geodesic error on the FAUST, SMAL, and TOSCA datasets, outperforming current mainstream algorithms in terms of computational efficiency and effectively addressing the issues of maintaining directional consistency and improving correspondence accuracy in non-rigid model matching.