Min/Max

Prev Next

The Min/Max Filter performs morphological operations on images. These are fundamental image processing operations useful for noise removal, edge detection, and mask/matte refinement.

Overview

Morphological operations use a "structuring element" (kernel) that slides across the image. At each position, the operation examines all pixels within the kernel and selects the minimum or maximum value.

Operations

Operation Description Use Case
Minimum (Erosion) Selects darkest pixel in kernel. Shrinks bright areas, expands dark areas. Remove small bright noise, shrink masks
Maximum (Dilation) Selects brightest pixel in kernel. Expands bright areas, shrinks dark areas. Fill small holes, expand masks
Open Min then Max. Removes small bright spots while preserving overall shape. Remove salt noise, smooth contours
Close Max then Min. Removes small dark spots while preserving overall shape. Remove pepper noise, fill small holes
Gradient Max minus Min. Highlights edges/boundaries. Edge detection, outline extraction

Settings

Operation

Setting Description
Operation The morphological operation to perform
Iterations Number of times to apply the operation (1-10). More = stronger effect.

Kernel

Setting Description
Radius X Horizontal radius of the kernel (1-100 pixels)
Radius Y Vertical radius of the kernel (1-100 pixels)
Link radii When ON, Radius Y follows Radius X for uniform kernel
Shape Shape of the structuring element

Kernel Shapes

Shape Description Best For
Box Square/rectangular kernel General purpose, fast
Ellipse Circular/elliptical kernel Isotropic operations, natural look
Cross Plus-shaped kernel (+) Directional operations
Diamond Diamond-shaped kernel (â—‡) Compromise between box and ellipse

Processing (Advanced)

Setting Description
Channel mode Per-Channel: process R, G, B independently. Luminance: select whole pixel by brightness.
Process channels Which channels to process (see below)

Process Channels

Option Description
RGB Process color channels only, preserve original alpha
RGBA Process all channels including alpha
Alpha Only Process only the alpha channel, preserve RGB colors

The Alpha Only option is particularly useful for:

  • Refining mattes and masks
  • Choking/spreading alpha edges without affecting color
  • Cleaning up keyed footage alpha

Keyboard Shortcuts

Key Function
F1 Cycle through operations
F2 Cycle through kernel shapes
F3 Toggle channel mode (Per-Channel/Luminance)
F4 Cycle process channels (RGB/RGBA/Alpha)
F5 Toggle link radii
R Reset all settings to defaults

Reset

Click the Reset all button or press R to restore all settings to their default values:

  • Operation: Maximum
  • Iterations: 1
  • Radius X/Y: 3
  • Link radii: On
  • Shape: Ellipse
  • Channel mode: Per-Channel
  • Process channels: RGB

Use Cases

Noise Removal

Salt noise (bright spots):

  • Use Open operation (Min then Max)
  • Small radius (1-3)
  • 1-2 iterations

Pepper noise (dark spots):

  • Use Close operation (Max then Min)
  • Small radius (1-3)
  • 1-2 iterations

Mask/Matte Refinement

Shrink mask (choke):

  • Use Minimum operation
  • Adjust radius to desired shrink amount
  • Use Ellipse shape for smooth edges

Expand mask (spread):

  • Use Maximum operation
  • Adjust radius to desired expansion
  • Use Ellipse shape for smooth edges

Clean up mask edges:

  • Use Open then Close (or vice versa)
  • Small radius (1-2)
  • Removes small artifacts

Edge Detection

  • Use Gradient operation
  • Adjust radius for edge thickness
  • Works like a simple edge detector

Creating Outlines

  1. Apply Maximum to expand the shape
  2. Subtract original from expanded version
  3. Result is an outline of the original shape

Channel Modes

Per-Channel Mode

Processes R, G, B channels independently. Each channel gets its own min/max calculation. This is the standard mode and works well for most cases.

Luminance Mode

Calculates brightness (luminance) for each pixel in the kernel and selects the entire pixel with the minimum/maximum luminance. This preserves color relationships better but may produce different results.

Performance Notes

  • Larger radii are slower (O(r²) complexity)
  • Box shape is slightly faster than Ellipse
  • Multiple iterations are slower than single iteration with larger radius, but may produce different results
  • Processing alpha adds minimal overhead

Comparison with Median Filter

Feature Min/Max Median
Noise removal Good for impulse noise Better for mixed noise
Edge preservation Can shift edges Better edge preservation
Speed Faster Slower
Compound ops Open/Close/Gradient N/A
Use case Mask refinement, morphology General denoising

Technical Details

The structuring element (kernel) determines which pixels are considered at each position:

Box 3x3:        Ellipse 3x3:    Cross 3x3:      Diamond 3x3:
■ ■ ■           · ■ ·           · ■ ·           · ■ ·
â–  â–  â–            â–  â–  â–            â–  â–  â–            â–  â–  â– 
■ ■ ■           · ■ ·           · ■ ·           · ■ ·

For non-square kernels (Radius X ≠ Radius Y), the shapes scale proportionally.