Engine Architecture

01

Input & Preprocessing

Images are normalized to Linear RGB space to prevent gamma-correction artifacts during color averaging. High-frequency noise is removed using bilateral filtering, preserving significant edges while smoothing textile weave textures.

> Bilateral Filter (sigma_s=2.0)
> Gamma Linearization (1.0)
> Histogram Equalization (Local)
02

Honeycomb Tessellation

Unlike grid-based pixels, we project a hexagonal honeycomb mesh onto the image. Hexagons have 6 neighbors (vs 4), allowing for smoother gradient interpolation and more organic boundary tracing.

03

Wedge Gradient Sampling

Each cell is split into 6 wedges. We sample color at the Cell Center and the Edge Midpoint. A linear gradient is constructed between these points. This creates a continuous color field that mimics the original photograph without posterization.

04

Residual Tile Fallback

For areas with chaotic high-frequency detail (like a rough wool knit) where vectors would be inefficient, the engine automatically generates "Residual Tiles"—high-compression raster patches that blend seamlessly with the vector output.