Degradr Effects Operations

Five operations provide optical, sensor, and analog media artifacts for creative image transformation. These operations are inspired by and adapted from degradr by nhauber99 (MIT License), migrated from PyTorch to SciPy/NumPy for PIL integration.

Quick Reference

Operation	Purpose	Key Parameters
blur_gaussian	Gaussian blur smoothing	sigma
blur_circular	Circular bokeh effects	radius
noise	Film grain and scan lines	mode, amount
chromatic_aberration	RGB channel fringing	shift_x, shift_y
bayer_filter	Digital sensor artifacts	pattern

Gaussian Blur

Apply Gaussian smoothing for dreamy softness and glow effects.

Basic Usage

- name: "soft_glow"
  operation: "blur_gaussian"
  params:
    sigma: 2.0

Parameters

• sigma (float): Standard deviation for Gaussian kernel
  • Range: 0.0+ (0 returns original image)
  • 0.5-2.0: Subtle blur, soft focus
  • 2.0-5.0: Moderate blur, dreamy glow
  • 5.0-10.0: Strong blur, heavy softness
  • 10.0+: Extreme blur, abstract smoothness

Algorithm

Uses scipy.ndimage.gaussian_filter with separable Gaussian kernels for efficient 2D convolution. Blur applied to RGB channels independently with alpha preserved.

Implementation Details:

• Edge handling: mode="reflect" to avoid dark borders
• Preserves original dtype (uint8)
• RGBA: Alpha channel unchanged, blur only on RGB
• Time complexity: O(n * k) where k = kernel size

Examples

Subtle Soft Focus:

- name: "gentle_glow"
  operation: "blur_gaussian"
  params:
    sigma: 1.0

Dreamy Aesthetic:

- name: "dream_blur"
  operation: "blur_gaussian"
  params:
    sigma: 4.5

Abstract Smoothness:

- name: "heavy_blur"
  operation: "blur_gaussian"
  params:
    sigma: 15.0

---

Circular Blur

Create lens-like circular bokeh effects simulating shallow depth of field.

Basic Usage

- name: "bokeh"
  operation: "blur_circular"
  params:
    radius: 8

Parameters

• radius (int): Radius of circular kernel in pixels
  • Range: 0+ (0 returns original image)
  • 3-8: Subtle bokeh, soft background
  • 8-15: Moderate bokeh, distinct circles
  • 15-30: Strong bokeh, pronounced effect
  • 30+: Extreme bokeh (computationally expensive)

Algorithm

Custom implementation using circular (disc-shaped) convolution kernel. Creates uniform averaging within a circular region.

Kernel Generation:

• Creates 2D grid centered at (0, 0)
• Circular mask: x² + y² ≤ radius²
• Normalized so sum equals 1 (preserves brightness)

Implementation Details:

• Applied via scipy.ndimage.convolve with mode="reflect"
• Each color channel processed independently
• RGBA: Preserves alpha, blurs RGB only
• Time complexity: O(n * r²) - expensive for large radii

Examples

Soft Background Blur:

- name: "subtle_bokeh"
  operation: "blur_circular"
  params:
    radius: 5

Lens Bokeh:

- name: "lens_bokeh"
  operation: "blur_circular"
  params:
    radius: 12

Dreamy Portrait:

- name: "portrait_blur"
  operation: "blur_circular"
  params:
    radius: 20

---

Noise

Add grain and scan line artifacts simulating film grain, sensor noise, and VHS degradation.

Basic Usage

- name: "film_grain"
  operation: "noise"
  params:
    mode: "gaussian"
    amount: 0.05
    seed: 42

Parameters

• mode (str): Noise distribution type
  • "gaussian": Pixel-level random noise (film grain)
  • "row": Horizontal scan line artifacts (VHS)
  • "column": Vertical line artifacts (glitch)
• amount (float): Noise intensity
  • Range: 0.0-1.0
  • Gaussian mode:
    • 0.02-0.05: Subtle film grain
    • 0.05-0.1: Moderate texture
    • 0.1-0.2: Heavy grain
    • 0.2+: Extreme noise
  • Row/Column mode:
    • 0.03-0.08: Subtle scan lines
    • 0.08-0.15: Moderate artifacts
    • 0.15-0.3: Strong lines
    • 0.3+: Extreme degradation
• seed (int, optional): Random seed for reproducibility

Algorithm

Three noise generation modes using NumPy random number generation:

1. Gaussian: rng.normal(loc=0, scale=amount, size=shape)
2. Row: rng.uniform(-amount, amount, size=(h, 1, c)) then broadcast
3. Column: rng.uniform(-amount, amount, size=(1, w, c)) then broadcast

Implementation Details:

• Image normalized to [0, 1] for processing
• Final clipping: np.clip(img + noise, 0.0, 1.0)
• Uses np.random.default_rng(seed) for modern RNG
• Handles grayscale, RGB, and RGBA

Examples

Film Grain:

- name: "analog_grain"
  operation: "noise"
  params:
    mode: "gaussian"
    amount: 0.06
    seed: 100

VHS Scan Lines:

- name: "vhs_scanlines"
  operation: "noise"
  params:
    mode: "row"
    amount: 0.12
    seed: 200

Glitch Lines:

- name: "vertical_glitch"
  operation: "noise"
  params:
    mode: "column"
    amount: 0.2
    seed: 300

---

Chromatic Aberration

Simulate lens chromatic aberration by shifting RGB channels independently, creating color fringing.

Basic Usage

- name: "color_fringe"
  operation: "chromatic_aberration"
  params:
    shift_x: 3
    shift_y: 0

Parameters

• shift_x (int): Horizontal shift in pixels
  • Range: -50 to 50 (typical)
  • Negative: Shifts left
  • Positive: Shifts right
  • Larger values = more pronounced fringing
• shift_y (int): Vertical shift in pixels
  • Range: -50 to 50 (typical)
  • Negative: Shifts up
  • Positive: Shifts down

Algorithm

Independent RGB channel shifting with green as reference:

• Red: (+shift_y, +shift_x) - shifts in specified direction
• Green: (0, 0) - reference, no shift
• Blue: (-shift_y, -shift_x) - shifts opposite direction

Implementation Details:

• Uses scipy.ndimage.shift for sub-pixel capability
• Shift mode: mode="constant", cval=0 (black fill)
• RGBA: Preserves alpha channel
• Non-RGB images: Returns copy without effect

Visual Effects

• Horizontal shift (shift_x only): Red-cyan fringing on vertical edges
• Vertical shift (shift_y only): Red-cyan fringing on horizontal edges
• Combined shift: Diagonal color separation
• Larger values: More pronounced optical distortion

Examples

Subtle Lens Aberration:

- name: "subtle_fringe"
  operation: "chromatic_aberration"
  params:
    shift_x: 2
    shift_y: 0

Horizontal Color Shift:

- name: "horizontal_aberration"
  operation: "chromatic_aberration"
  params:
    shift_x: 8
    shift_y: 0

Diagonal Distortion:

- name: "diagonal_shift"
  operation: "chromatic_aberration"
  params:
    shift_x: 5
    shift_y: 5

Extreme Glitch:

- name: "glitch_aberration"
  operation: "chromatic_aberration"
  params:
    shift_x: 20
    shift_y: -10

---

Bayer Filter

Simulate digital camera sensor artifacts by applying Bayer Color Filter Array mosaicing and demosaicing.

Basic Usage

- name: "sensor_artifacts"
  operation: "bayer_filter"
  params:
    pattern: "RGGB"

Parameters

• pattern (str): Bayer pattern arrangement
  • Options: "RGGB" (default), "BGGR", "GRBG", "GBRG"
  • Defines 2x2 pixel block color arrangement

Bayer Patterns

RGGB:  R G    BGGR:  B G    GRBG:  G R    GBRG:  G B
       G B           G R           B G           R G

Algorithm

Two-stage process:

1. Mosaicing: Convert RGB to Bayer CFA (single channel)
2. Demosaicing: Reconstruct RGB using Malvar2004 algorithm

Implementation Details:

• Uses colour-demosaicing library
• Algorithm: demosaicing_CFA_Bayer_Malvar2004
• High-quality edge-aware interpolation
• RGBA: Preserves alpha, effect on RGB only

Visual Artifacts

• Color fringing at edges
• Moiré patterns on fine details
• Slight softening from interpolation
• Pattern-dependent color shifts

Examples

Standard Sensor:

- name: "digital_sensor"
  operation: "bayer_filter"
  params:
    pattern: "RGGB"

Alternative Pattern:

- name: "bggr_sensor"
  operation: "bayer_filter"
  params:
    pattern: "BGGR"

---

Combining Degradr Operations

Create complex aesthetic pipelines by layering operations.

VHS Tape Aesthetic

pipeline:
  steps:
    - name: "color_shift"
      operation: "chromatic_aberration"
      params:
        shift_x: 3
        shift_y: 0

    - name: "scanlines"
      operation: "noise"
      params:
        mode: "row"
        amount: 0.08

    - name: "vhs_blur"
      operation: "blur_gaussian"
      params:
        sigma: 1.5

    - name: "vhs_compress"
      operation: "compression"
      params:
        quality: 35
        subsampling: 2

Digital Camera Simulation

pipeline:
  steps:
    - name: "sensor_mosaic"
      operation: "bayer_filter"
      params:
        pattern: "RGGB"

    - name: "sensor_noise"
      operation: "noise"
      params:
        mode: "gaussian"
        amount: 0.04

    - name: "camera_compress"
      operation: "compression"
      params:
        quality: 85
        subsampling: 2

Vintage Lens Character

pipeline:
  steps:
    - name: "bokeh"
      operation: "blur_circular"
      params:
        radius: 10

    - name: "lens_aberration"
      operation: "chromatic_aberration"
      params:
        shift_x: 4
        shift_y: 2

    - name: "boost_colors"
      operation: "saturation"
      params:
        mode: "fixed"
        value: 1.3

---

Performance Characteristics

Computational Cost

Sorted by typical performance (fastest to slowest):

1. Chromatic Aberration: O(n) - simple channel shifts
2. Noise (Row/Column): O(n) - broadcasting
3. Noise (Gaussian): O(n) - per-pixel RNG
4. Gaussian Blur: O(n * k) - separable convolution
5. Circular Blur: O(n * r²) - 2D convolution
6. Bayer Filter: O(n * k) - demosaicing interpolation

Memory Usage

• Gaussian Blur: Minimal (in-place capable)
• Circular Blur: O(r²) for kernel
• Noise: O(n) for noise array
• Chromatic Aberration: O(n) for output
• Bayer Filter: O(2n) for mosaic + demosaiced

Optimization Tips

• Use smaller radii/sigma for faster processing
• Noise operations benefit from fixed seed for consistency
• Consider downscaling for preview generation
• Batch process frames when possible

---

Attribution

Original Implementation: degradr by nhauber99 (MIT License)

Key Adaptations:

• Migrated from PyTorch to SciPy/NumPy
• Added comprehensive parameter validation
• Implemented consistent RGBA alpha handling
• Optimized for frame-by-frame video processing

License: See LICENSE_DEGRADR.txt in repository

---

Next Steps

• Try hands-on examples: Degradr Effects Tutorial
• Explore all parameters: Filter Guide
• Learn YAML system: Pipeline Documentation
• Technical details: Full algorithm documentation in repository technical notes

---

For comprehensive parameter ranges and visual examples, see the complete Filter Guide.