Reliable Cosine Matching via Neighborhood Consensus for Fundus Image Registration

Fundus image registration holds significant importance in medical image analysis. This research addresses the challenges inherent in fundus image registration through the introduction of a novel framework, denoted as Reliable Cosine Matching via Neighborhood Consensus (RCM-NC). The framework aims to enhance the precision and reliability of automatic fundus image registration by refining both the pre-establishment of pairing relations and the removal of mismatches. Key innovations include the incorporation of a Nearest Neighbor Distance Ratio Extension (NNDRE) to broaden the spectrum of initial matches, while simultaneously preserving the accuracy of matching outcomes. Furthermore, Reliable Cosine Matching (RCM) strategies leverage neighborhood consensus to scrutinize and validate matching pairs, improving the overall robustness of the registration process. Empirical assessments on the public fundus image registration dataset demonstrate the superior performance of the proposed method compared to traditional and deep learning approaches.