If you run a simple edge detector or generic AI image filter on a modern Full Art or Secret Rare Pokémon card, it goes haywire. It will flag 4,000 "micro-scratches" across Pikachu's cheek.
Those aren't scratches. That is the diffraction grating of the holographic foil and microscopic halftone ink dots reflecting light.
The Problem: Artwork Noise vs. Physical Scratches
On a raw image, pixel intensity changes rapidly wherever glitter catches the light. To a naive computer program, a sharp contrast jump looks identical to an edge:
- Holo Glitter: Tiny circular bursts of light ($10\text{–}30\,\mu\text{m}$ in diameter).
- Real Scratch: A continuous 1D furrow plowed into the protective clear coat, often spanning $2\text{ to }15\text{ mm}$ in length.
"Directional derivatives of any orientation can be synthesized by a linear combination of a small basis set. By decomposing local intensity curvature via the Hessian matrix, true curvilinear ridges are enhanced independently of background noise."
Steering Filters to the Angle of the Scratch
In 1991, MIT researchers William Freeman and Edward Adelson discovered that you don't need to rotate thousands of filters to find an angled line. Instead, by taking just two orthogonal basis derivatives ($I_x$ and $I_y$), you can mathematically "steer" a directional filter to any angle $\theta$:
GradieQ applies this to construct an anisotropy ratio:
- Isotropic texture (glitter & rosettes): Energy is equal in all directions. The filter cancels it out (over 85% suppression).
- Anisotropic defect (real scratch): Energy peaks sharply at the scratch's specific orientation angle $\theta$. The signal stands out with high contrast.
That is why GradieQ can inspect high-end vintage and modern alternate arts without misinterpreting the card's natural holographic pattern as surface wear.