Currently only Pokémon cards can be verified — other card games are not supported yet.
Edge Metrology

The 0.05 mm Difference: Why Subpixel Math Decides a PSA 10

How thick is a single pixel? On a standard 12-megapixel smartphone photo of a trading card, 1 pixel represents approximately 0.06 mm (60 micrometers) of physical card surface.

Now consider the grading guidelines:

  • PSA 10 Gem Mint requirement: Front centering between $50/50$ and $60/40$.
  • PSA 9 Mint: Centering worse than $60/40$ (e.g. $61/39$ or $62/38$).

On a standard Pokémon card margin ($w \approx 3.2\text{ mm}$), the difference between a $59/41$ (PSA 10) and a $61/39$ (PSA 9) is less than 0.12 mm — literally about two camera pixels.

The Danger of Discrete Rounding

If a grading app simply counts integer pixels, rounding error is $\pm 0.5\text{ px}$. That single half-pixel rounding can swing a card's calculated centering from $59.5/40.5$ (eligible for 10) to $60.5/39.5$ (downgraded to 9).

Relying on integer pixels turns high-end card verification into a coin toss.

"An edge is marked at the local maximum of the gradient magnitude. By fitting a continuous second-degree polynomial across discrete gradient samples, the true edge location can be localized to a fraction of a pixel with sub-micron repeatability."
— Frédéric Devernay & Carsten Steger, Sub-Pixel Edge & Line Metrology (INRIA / IEEE TPAMI)

3-Point Parabolic Subpixel Localization

Instead of snapping to the nearest pixel grid, GradieQ casts 128 rays across every card margin and computes the continuous first derivative of a Gaussian smoothing filter:

$$f(t) = \frac{\partial}{\partial t} (G_\sigma * I)(t)$$

Let $t_0$ be the discrete pixel with the highest gradient. The true subpixel peak $t^*$ lies between pixels and is solved via continuous 3-point parabolic interpolation:

$$t^* = t_0 + \frac{f(t_0 - 1) - f(t_0 + 1)}{2 [f(t_0 - 1) - 2f(t_0) + f(t_0 + 1)]}$$

This technique achieves edge localization repeatability down to $\pm 0.05\text{ mm}$ ($\pm 5\,\mu\text{m}$) across 512 independent points per face. That is why GradieQ's centering ratios are steady, repeatable, and immune to sensor grid jitter.

Verify Your Card with Deterministic Optics

Upload a standard photo to measure centering and corners and see what the photo can and can’t tell.

Grade a card now — Free Open Beta