New Paper: Selective Disk Bispectrum, a Complete Rotation-Invariant Image Descriptor
ninamiolane · x · 2026-11-21
Nina Miolane shared the team's new arXiv paper, Selective Disk Bispectrum: A Complete and Rotation Invariant Image Descriptor by Adele Myers and Nina Miolane.
- Proposes the selective disk bispectrum (SDB), a complex-valued rotation-invariant vector that preserves all image information except orientation, bridging hand-crafted invariant features and learned representations
- Theoretical contributions include invertibility, reduced spatial/computational complexity vs. the full disk bispectrum, and noise expectation/variance analysis
- A numerical approximation comes with accuracy and rotation-invariance guarantees; experiments validate invariance, noise robustness, and reconstruction on multi-reference alignment of rotated images
Related prior work: Bispectral Neural Networks and The Selective G-Bispectrum.
More from Research
- William Hoza Proves NL = UL in Strikingly Simple Log-Space Construction — fortnow · 2026-11-21
- Why the Riemann Hypothesis is hard: zeta on the 4/5-line gets arbitrarily close to zero — AlexKontorovich · 2026-11-21
- Stanford Study: Deactivating AI for Two Weeks Reduces Loneliness and Boosts Social Interaction — stanfordnlp · 2026-11-21
- DOCKET Benchmark: Only 18-30% of LLM Policy Comments Align With Real Human Feedback — stanfordnlp · 2026-11-21
- Stanford AI100 releases 2026 report on the scaling era and mainstream generative AI — stanfordnlp · 2026-11-21
- New Multi-Agent Coordination Bench finds frontier agents coordinate poorly; only Opus gains — AlexGDimakis · 2026-11-21