GIMP Waterpixels Compactness and Superpixel Regularity

This article explains how the compactness parameter functions within the GIMP Waterpixels filter to dictate the geometric regularity of generated superpixels. By reading further, you will understand the mathematical and spatial balance between edge adherence and shape uniformity, how adjusting this setting alters superpixel geometry, and how to choose the right value for tasks ranging from precise image segmentation to artistic stylization.

The Role of Waterpixels in GIMP

The Waterpixels filter in GIMP generates superpixels—clusters of adjacent pixels grouped by shared characteristics—using a spatial-regularized adaptation of the watershed transform. Standard watershed segmentation identifies local minima in an image's gradient and expands "water basins" outward until they collide, forming boundaries. While effective at finding strong edges, standard watershed algorithms often yield erratic, arbitrarily shaped regions. The Waterpixels algorithm introduces a regular grid of markers and incorporates spatial constraints to keep the resulting segments manageable, cohesive, and uniform.

How Compactness Influences Superpixel Geometry

The compactness parameter controls the relative weight between two opposing forces: color/gradient distance and spatial distance.

When water basins expand from their initial seed points, the algorithm calculates a cost function for absorbing neighboring pixels. This cost combines:

  1. Photometric cost: The difference in color intensity or gradient value between pixels.
  2. Spatial cost: The physical Euclidean distance from the pixel to the superpixel's seed center.

The compactness slider directly scales the spatial cost.

Low Compactness: High Boundary Adherence, Low Regularity

When compactness is set to a low value, the spatial constraint is minimized. The algorithm prioritizes color gradients and local edges above all else:

  • Shape Dynamics: Superpixels flow freely into complex, organic contours, bending and stretching to follow natural borders in the image.
  • Geometric Regularity: Regularity is extremely low. Superpixels will vary wildly in size, aspect ratio, and convexity, often producing elongated or jagged patches.
  • Trade-off: High boundary recall at the expense of geometric stability.

High Compactness: Low Boundary Adherence, High Regularity

When compactness is increased, spatial distance is prioritized over weak or moderate color differences:

  • Shape Dynamics: The algorithm penalizes pixels that lie far from their seed markers. Basins grow at more uniform speeds in every direction regardless of underlying texture.
  • Geometric Regularity: Regularity is very high. Superpixels become nearly uniform in area and assume compact, convex shapes that resemble a regular hexagonal lattice or Voronoi tessellation.
  • Trade-off: Strong geometric uniformity at the expense of boundary precision, often causing superpixels to bleed across subtle image edges.

Practical Guidelines for Adjusting Compactness

  • For Pre-Processing and Object Segmentation (Low to Moderate Compactness): Keep compactness low (around 10% to 30%). This allows the superpixels to wrap tightly around subject boundaries, making them ideal for subsequent masking, feature extraction, or machine vision workflows.
  • For Stylization and Mosaic Effects (High Compactness): Set compactness higher (around 60% to 90%). This enforces a structured, tile-like aesthetic where the geometric layout remains visually consistent and visually calm across varying image textures.