Non-Learning Low-Light Stereo Vision

2026-05-29

AI Summary

This work addresses the challenge of disparity estimation from extremely low-light, high-noise stereo images by proposing a learning-free stereo matching framework. The method constructs a cost volume using coarse-grained structural features derived from the Field of Junctions (FoJ), which exhibit robustness to noise. A boundary-aware semi-global matching (SGM) strategy is introduced to dynamically adjust smoothness penalties, thereby preserving true disparity discontinuities. By eschewing deep learning and relying solely on geometric structure cues, the approach achieves high-accuracy sparse disparity estimation under severe noise conditions, effectively distinguishing genuine edges from photon noise. Experimental results demonstrate that, on standard benchmark datasets, the proposed method outperforms state-of-the-art stereo algorithms in sparse disparity accuracy within non-occluded regions.

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.

Problem

• low-light stereo vision
• disparity estimation
• severely noisy images
• stereo matching

Innovation

• Non-learning stereo
• Low-light vision
• Field of Junctions
• Boundary-aware SGM
• Disparity estimation