基于无监督学习的风洞压敏漆图像配准算法

Abstract

structure and a coarsetofine registration strategy to address the issue of large offsets and nonrigid deformations in wind tunnel environments. Additionally, it incorporates a novel loss function paradigm based on image similarity, which enhances image registration in featuresparse scenarios. Comparing with typical alignment methods such as MI -Spline and Voxelmorph on two sets of PSP images, the experimental results prove that our method achieves a far better registration performance than the existing methods in visual evaluation and RMSE, NCC, and TRE metrics, under the premise of ensuring the performance of the algorithm. This method provides a reliable solution to the PSP image registration problem .