Twelve masks. Twelve theories of what structure means in an image, from luminance edges to how the eye actually sees color. Upload one. The scatter plot shows where they agree and where they don't.
original
where does each mask say the image lives?
primitive definitions
ΔxHorizontal centroid offset. Where the gradient mass sits left or right of frame center. Negative is left.
ΔyVertical centroid offset. Where the gradient mass sits above or below frame center. Negative is up.
rᵥVoid ratio. Fraction of the image with near-zero gradient response. High rᵥ means most of the frame is empty — only a small region carries structure.
μCohesion. Fraction of total gradient energy concentrated in the top-quartile pixels. High μ means the mass is tight. Low μ means it is scattered thin.
SDISpatial Dispersion Index. Mean distance of mass pixels from the centroid, normalized by the image diagonal. Low is a center blob. High is edge-weighted spread. Detects collapse into center monoculture even when rᵥ and μ look stable.
ρᵣPacking density. Mass area relative to convex hull area, ×100. How tightly the active marks are compressed within their envelope. Low means marks are sparse inside their bounding shape.
xₚPeripheral edge pull. Fraction of gradient energy in the outer 15% band of the frame. High means mass lives at the edges. Low means centripetal collapse.
θOrientation stability. One minus the Shannon entropy of the 8-bin gradient orientation histogram, weighted by magnitude. High means edges are directionally consistent. Low means omnidirectional noise.
dₛStructural thickness. Mean ridge width at active pixels, normalized by the shorter image dimension. Thin edges score low. Broad masses score high.
baseline (Sobel / LAB-L)
confirmed (VTL × L−M)