CAISOV: Collinear Affine Invariance and Scale-Orientation Voting for Reliable Feature Matching

Reliable feature matching plays an important role in the fields of computer vision and photogrammetry. Due to complex transformation model caused by photometric geometric deformations, limited discriminative power local descriptors, initial matches with high outlier ratios cannot be addressed very well. This study proposes a reliable outlier-removal algorithm combining two affine-invariant constraints. First, simple constraint, namely, CAI (collinear affine invariance) has been implemented, which is based on observation that collinear property any points invariant under transformation. Second, after first-step removal SOV (scale-orientation voting) scheme was then adopted remove remaining outliers recover lost inliers, peaks both scale orientation voting define parameters model. Finally, match expansion executed using Delaunay triangulation refined matches. By close-range (rigid non-rigid images) UAV (unmanned aerial vehicle) datasets, comprehensive comparison analysis are conducted this study. The results demonstrate proposed achieves best overall performance when compared RANSAC-like constraint-based methods, it can also applied achieve workflow SfM-based image orientation.