Constant-Time Three-Dimensional Integral Histograms and Fixed-Point Lookup-Table Spherical Entropy for Real-Time Flow Diagnostics
Year: 2026
Type: preprint
Abstract: This work introduces a strictly division-free, square-root-free, and logarithm-free fixed-point spherical entropy descriptor with provable deterministic accuracy and perturbation guarantees, designed for FPGA, ASIC, and GPU pipelines. The formulation completely eliminates divisions, square roots, and atan2 in spherical binning through exact cross-product sector tests and quadratic axial comparisons on an equal-area Lambert–Archimedes partition, while preserving full bin-index equivalence. It extends two-dimensional summed-area tables and integral histograms to three dimensions with spatial tiling, so that window queries remain constant-time with respect to radius and fit within bounded on-chip memory. Runtime divisions and logarithms are replaced by an integer lookup table that guarantees a deterministic and bounded quantization error. In addition, a tight local robustness bound against sensor dropouts is derived from the Fannes–Audenaert continuity inequality, with no violations observed in practice. The descriptor is evaluated alongside classical gradient criteria and three-dimensional phase-unwrapping techniques across regimes of extreme aliasing, low-SNR signal voids, and shear layers, clarifying its role as a complementary directional-disorder measure with deterministic guarantees.
Author: Porolo, Andres
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7233602416
Cites:
Cited by:
Related to:
