OffTheDop
✦ Spatial Sobel Optics & 19-Tilt Scintillation 8 min read Explore High-Contrast Cuts →

MAKING CONTRAST COMPUTABLE: SOBEL GRADIENTS, DENSITY VS. INTENSITY, AND THE ASET BLUE MYTH

• ProFacet Computational Faceting
Physics & Optics Computation

A quick note: I’m not a trained gemologist. No GIA diploma, no mineralogy degree, and no traditional bench apprenticeship. Just a software engineer who fell down the rabbit hole of faceting machines, geometry, and how light bounces around inside crystals.

Ask three experienced gem cutters what gives a finished stone "life" on the hand, and they will reach for words from photography and music: pop, crispness, snap, checkerboard flash, crushed ice. Every cutter recognizes contrast the instant a stone tilts under room lighting. Without dark reflections of the observer's head and shoulders interrupting bright reflections of the overhead sky, even complete light return looks like a flat white bathroom mirror.

Yet for decades, contrast in gemstone design remained either an unquantified artistic vibe or a crude single-frame shortcut: counting how much blue ink appeared inside a static face-up ASET diagram. You cannot feed an artistic vibe into a GPU optimizer evaluating thousands of candidate polyhedra per second. And as our catalog data shows, optimizing for static ASET blue area actually steers an algorithm straight into central ink-puddle extinction.

To build ProFacet and OffTheDop, we had to make spatial contrast mathematically computable across real 3D tilt, and split it into two independent physical dimensions that traditional gemology never separated: how finely light and shadow are woven across the crystal (Contrast Density) versus how steeply each bright-to-dark cliff drops (Contrast Intensity).

Path-traced Showcase Moissanite mirrored-truncated trilliant Luminous Ridge 664 exhibiting razor-sharp dark-to-bright scintillation and rainbow dispersion
Luminous Ridge 664 (Showcase Collection, Moissanite mirrored trilliant). High light return paired with crisp, edge-to-edge 3D scintillation that holds steady across everyday hand motion.

From Subjective Pop To A 19-angle Gpu Sobel Operator

If you want a computer to measure contrast the way a human eye experiences scintillation, you have to model two physical realities at once: spatial edge detection across neighboring pixels on the retina, and motion as the stone rocks back and forth on a finger or pendant.

Inside ProFacet's WebGPU optical analyzer (local_contrast_atlas.wgsl), every candidate gemstone runs through a four-stage spatial gradient pipeline in less than a millisecond:

  1. A multi-angle raytracing atlas: Rather than judging a stone frozen face-up, the GPU raytraces 19 tilt frames sweeping from -45° to +45° under a realistic lighting hemisphere that includes the dark viewer head-shadow cone, mid-angle ambient sky, and bright overhead illumination.
  2. Perceptual luminance and rim erosion: Each pixel's RGB return is converted into human photopic luminance, and the outer gemstone silhouette is eroded inward by one pixel. Without that rim erosion, any bright stone sitting against a black background gets rewarded with a fake ring of free contrast around its girdle perimeter, biasing an optimizer toward jagged outlines.
  3. Sobel spatial gradient convolution: At every interior pixel, the shader slides horizontal and vertical 3x3 Sobel derivative kernels across its eight neighbors to measure local edge gradient magnitude. Smooth, uniform slabs of light evaluate near zero; sharp facet-to-facet jumps between overhead sky and dark head-shadow light up as crisp ridges.
  4. Separating Density from Intensity: Averaging edge strength blindly across a stone confuses a dense grey silk pattern with two giant black-and-white bars. Instead, ProFacet splits the gradient map into two distinct signals: Contrast Density (the share of interior crystal pixels carrying an active optical boundary, measuring how finely light and shadow interleave) and Contrast Intensity (the average gradient steepness strictly along those active boundaries, measuring how sharply each cliff snaps from bright to dark).
Three-stage engine-computed GPU pipeline showing the real 19-tilt plus ASET raytraced atlas, Rec. 709 luminance with 1-pixel rim erosion, and 3x3 Sobel gradient edge decomposition on Diamond Hearts and Arrows
Real GPU analyzer pass on classic Diamond Hearts & Arrows. Stage 1 raytraces all 19 tilt frames plus ASET in one atlas. Stage 2 extracts perceptual luminance inside an eroded rim mask. Stage 3 runs 3x3 Sobel edge detection to separate boundary weave (Contrast Density) from boundary cliff steepness (Contrast Intensity).

Across the tilt sweep, ProFacet weights angles near face-up highest while still demanding strong retention out to 45°, normalizes both metrics against a classic Diamond Hearts & Arrows round brilliant (set to 1.00x), and applies a smooth diminishing-returns curve above unity so an optimizer never sacrifices brightness just to chase microscopic aliasing.

Why Static 0-degree Aset Blue Fails To Predict Contrast

Anyone who has shared a faceting diagram on an online gemology forum has met the ASET Blue rule of thumb. In an ASET map, red marks overhead light, green marks low-angle ambient light, and blue marks the narrow cone blocked by the observer's head and camera lens. Because viewer head-shadow is the raw ingredient that creates dark contrast facets in a diamond, conventional wisdom jumped to a convenient shortcut: just measure the percentage of blue pixels in a static face-up ASET diagram to grade contrast.

We tested that shortcut directly against 3D multi-angle Sobel contrast across nearly 4,800 catalog designs sampled evenly across all eight refractive indices, from Quartz up to Moissanite.

The result is sobering: static face-up ASET Blue percentage has almost zero ability to predict real 3D contrast (correlating at r = -0.17), and what slight trend exists actually points in the wrong direction:

Empirical scatter plots across 4,799 OffTheDop gemstones demonstrating why 0-degree ASET Blue fails to predict 3D contrast and mapping the four orthogonal quadrants of Contrast Density versus Contrast Intensity
Left: Across 4,799 designs spanning all eight refractive indices, static face-up ASET Blue percentage fails to predict 3D dynamic contrast because large central head-shadow puddles inflate area while starving edge boundaries. Right: Contrast Density and Contrast Intensity are nearly independent, opening four distinct optical personalities.

Counting face-up blue ink fails because area is not perimeter, and a frozen snapshot is not 3D motion:

Four Contrast Personalities: Wisely Chosen Examples From The Catalog

As the right half of Figure 2 shows, Contrast Density and Contrast Intensity are nearly independent across our catalog. On OffTheDop, we combine them into a single headline Contrast score weighted predominantly toward spatial weave with a smaller bonus for edge sharpness.

Because those two underlying axes are independent, two gemstones with the exact same headline Contrast score can look remarkably different in person, while lower-density architectural cuts can still deliver striking terrace drama. Here are four curated Showcase and Signature designs drawn from four corners of that space:

Compare the top pair in that gallery, Ethereal Eclipse 308 and Flawless Echo 492. Both are RI 2.15 trilliants with identical brightness and the exact same 1.08x combined Contrast score, yet any jeweler holding them side by side immediately sees two distinct personalities:

The same contrast mechanics scale cleanly from low-RI Quartz up to Diamond:

Here is how those seven designs compare against the Diamond Hearts & Arrows baseline (1.00x):

Design Material Brightness Density Intensity Combined Contrast Face-Up ASET Blue Visual Personality
Hearts & Arrows (Round ref.) Diamond 2.42 1.00x 1.00x 1.00x 1.00x 22% Classic radial benchmark
Luminous Ridge 664 Moissanite 2.67 1.07x 1.14x 1.12x 1.14x 25% Electric weave and steep cliffs
Flawless Enigma 461 Diamond 2.42 1.06x 1.15x 1.13x 1.15x 22% Frontier diamond scintillator
Zenith Drift 567 CZ 2.15 0.93x 1.16x 1.07x 1.16x 19% Bowtie-free crushed-fire oval
Ethereal Eclipse 308 CZ 2.15 1.04x 1.11x 0.85x 1.08x 21% Dense, velvety woven silk
Flawless Echo 492 CZ 2.15 1.04x 1.07x 1.20x 1.08x 28% Bold graphic piano-key cliffs
Crystal Glacier 961 CZ 2.15 0.73x 0.91x 1.15x 0.93x 4.5% Crisp architectural step terraces
Iridescent Vortex 571 Quartz 1.54 1.10x 1.00x 1.02x 1.00x 9.7% Diamond-level contrast in Quartz

Tilt Stability And Steering Contrast In Profacet And Offthedop

Why evaluate Sobel gradients across 19 tilt angles instead of running edge detection once on the face-up image? Plotting real optical retention across the full viewing arc makes the difference obvious:

19-angle dynamic tilt performance curves from minus 45 to plus 45 degrees comparing high-stability cuts against a 0-degree face-up trap alongside ProFacet optimizer presets
Left: Multi-angle optical retention from -45° to +45°. Designs optimized across tilt (Luminous Ridge 664, Zenith Drift 567, and Iridescent Vortex 571) maintain contrast across everyday hand motion, whereas a face-up-only step cut (Crimson Serenade 499) collapses once tilted. Right: Optimizer weight presets in ProFacet Studio.

A cut tuned strictly for a vertical camera looks respectable dead-center, then opens a wide tilt window the moment a wearer rotates their wrist. Multi-angle Sobel scoring keeps contrast alive across the entire gesture.

Both platforms put these controls directly in your hands:

Even with multi-angle Density and Intensity in place, our GPU optimizer uncovered one more surprise. Occasionally a cut scored high on both Brightness and Sobel Contrast, yet rendered like a stark black-and-white barcode with almost no silvery midtones in between. Solving that blind spot meant bringing Claude Shannon's information theory into gemstone optics, which is the story of our companion post: All Thirty-Two Shades of Grey: Bringing Shannon Entropy to Gemstone Design.

19-Tilt WebGPU Sobel Scorer · Density & Intensity Controls

Steer Contrast Live in ProFacet or Read Companion Post on Shannon Entropy

Experiment with Contrast Density and Contrast Intensity sliders in ProFacet Studio, filter 92,000+ path-traced cuts by 3D Contrast on OffTheDop, or continue to Part 2 on 32-bin Claude Shannon Information Entropy.