Title: Non-Learning Low-Light Stereo Vision

Original: arXiv:2606.00379v1 Announce Type: new Abstract: We present a non-learning stereo framework for disparity estimation from severely noisy images. Using the Field of Junctions (FoJ), it retains coarse visual features stable under severe noise for cost volume construction while discarding fine textures inseparable from photon noise. The resulting structural information guides boundary-aware Semi-Global Matching (SGM) that dynamically adapts smoothness penalties to preserve true disparity discontinuities. The output is a sparse disparity map more accurate than those of recent stereo algorithms over unmasked pixels on widely-used benchmark datasets.

Rewrite: We introduce a novel, learning-free stereo approach designed to estimate disparity from images plagued by significant noise. By leveraging the Field of Junctions (FoJ), the method stabilizes coarse visual features essential for constructing cost volumes, effectively filtering out fine textures that are indistinguishable from photon noise. This extracted structural data informs a boundary-aware Semi-Global Matching (SGM) process, which adjusts smoothness penalties on the fly to maintain genuine disparity edges. Consequently, the system generates a sparse disparity map that demonstrates superior accuracy compared to contemporary stereo algorithms when evaluated on unmasked pixels across standard benchmark datasets.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC