Grounded in science

A method for producing dense local structural masks from sliding-window statistics — a continuous map of where structural concentration and spatial variability are locally significant. Fully unsupervised and data-driven: no training, no prior annotations.

The masks stay stable across a wide range of natural images and acquisition conditions, and capture meaningful structure beyond classical edge- or gradient-based representations — a generic guidance signal for downstream tasks.

An unsupervised way to extract regions of interest from local structural organization alone — no semantic interpretation, no prior knowledge of scene content. Dense structural masks are aggregated into coherent zones that concentrate a significant share of an image's structure.

On 120 natural images (Flickr2K) and the UETT4k-Anti-UAV set (30 images), it delivers 95–99% spatial reduction, outperforms Sobel/Laplacian baselines and matches Shannon-entropy approaches — with higher spatial coherence.

A multi-scale formulation of the structural measure φ built on inter-scale consensus: responses are computed across several spatial scales and combined by a vote-based mechanism, removing the sensitivity of single-scale masks to window size.

The multi-scale version yields more compact masks, fewer spurious regions and significantly more stable ROI selection across window sizes than single-scale formulations — a robust canonical extension.

A super-resolution approach that conditions the network explicitly on a local structural measure Φ, computed deterministically from the low-resolution input alongside luminance and chrominance. The explicit prior drives spatially adaptive reconstruction of fine detail while staying conservative in flat regions — limiting uncontrolled hallucination.

On a single-step ×4 model over 223 Flickr2K images, Φ-conditioning consistently improves reconstruction over classical interpolation baselines, acting as an interpretable structural prior — most of all in regions of high structural complexity.

Enhancement confined strictly to automatically detected structural regions of interest — nothing is touched outside them. Two mechanisms are studied: structure-guided unsharp masking, and CLAHE with a structurally adaptive clip limit.

Across 200 natural images, 15 methods and 4 window sizes, the multi-scale φ-guided CLAHE variant beats every gradient- and entropy-based baseline: highest in-ROI enhancement with the lowest leakage outside, and at its optimal scale it surpasses standard CLAHE by 73%.

An executive summary of the quality characterization of PHI-SR, a continuous-zoom super-resolution pipeline with arbitrary scale factors (typically ×2 to ×8). It consolidates three axes: stability of the φ signal under degradation, robustness on real sensor captures, and cross-domain portability (photography, urban, aerial, thermal infrared).

The φ signal keeps a positive correlation to its reference across every degradation regime; on real Canon/Nikon captures PHI-SR shows a perceptual advantage (LPIPS) over LIIF — including under added degradation, and markedly so on thermal infrared.

A preliminary note on the effect of a lightweight, proprietary modification of the input stream of a pixel-native JEPA world model — deterministic, local and low-cost, applied without changing the task, architecture or training. Only relative deltas and aggregate success rates are reported.

Across two visuomotor environments: up to −75% prediction loss on clean data; planning success under noise lifted from 28% to 44%; on noisy manipulation the baseline rises from 0% to 4–8%. The clean-manipulation effect stays inconclusive.