面向非刚性形变图像的关键点匹配方法OA

Non-rigid deformations significantly impair the performance of traditional keypoint matching algorithms in image matching tasks,re-sulting in reduced matching accuracy.To address this issue,a matching method based on probabilistic modeling and feature enhancement is proposed.First,an adaptive keypoint sampling strategy is designed to resolve the redundancy in keypoint distribution under complex deforma-tions by employing dynamic spatial constraints and boundary-aware filtering.Second,to overcome the limitation where deterministic geometric transformation estimation tends to fall into local optima,a Student-t distribution is introduced to model the uncertainty of geometric transforma-tion parameters,thereby enhancing descriptor adaptability in complex deformation scenarios.Finally,a feature enhancement module based on Gaussian priors is constructed.This module utilizes direction-sensitive operators to reinforce key structural information,effectively mitigating feature drift caused by non-rigid deformations.Experimental results demonstrate that the proposed method effectively improves matching accura-cy for non-rigid images and exhibits superior robustness in scenarios involving complex non-rigid deformations.