RRSI descriptor enables radiation, rotation and scale-invariant multimodal image matching

zhenjun_zhao · x · 2026-09-11

A new arXiv paper proposes a Radiation, Rotation and Scale Invariant (RRSI) descriptor for multimodal image matching. A dual-head regional sampling module performs Cartesian and Log-Polar sampling on keypoint neighborhoods; geometric and radiometric relations are jointly encoded in one deep feature space; and a bidirectional cross-modal generative reconstruction constraint anchors modality-invariant topology with zero extra inference cost. It is competitive on optical-infrared and optical-SAR datasets, supporting full 0–360° rotation and scale changes.

Original post →

More from Research

Research channel →