Phase Congruence Detector Parameters

Core Detection Parameters

Scales (Detection Scales)

• Default: 4
• Range: 2-8
• Description: Number of scale levels for multi-scale feature detection. Each scale detects features of different sizes.
• Guidelines:
  • Lower values (2-3): Faster processing, detects features of similar size, good for uniform textures
  • Higher values (5-8): Slower but more robust, detects features across a wider size range, better for natural images with varying feature scales
• Recommendation: Use 4 for general purpose, increase to 6-8 for images with both fine details and large structures

---

Orientations (Orientation Filters)

• Default: 8
• Range: 4-16
• Description: Number of orientation filters covering 180 degrees. Determines angular resolution of edge detection.
• Guidelines:
  • Lower values (4-6): Faster, coarser angular resolution (45-30° steps), suitable for detecting main structural directions
  • Higher values (12-16): Finer angular discrimination (15-11° steps), better for curved or complex features
• Recommendation: Use 8 for balanced performance, increase to 12-16 for images with many different edge orientations

---

Min Wavelength (Minimum Feature Size)

• Default: 3.0 pixels
• Range: 2.0-10.0
• Description: Smallest feature size (in pixels) that will be detected. Sets the highest frequency filter.
• Guidelines:
  • Lower values (2-3): Detects very fine details and texture, may increase noise sensitivity
  • Higher values (5-10): Ignores fine details, focuses on larger structural features, more noise-resistant
• Recommendation: Set to slightly smaller than your smallest feature of interest; 3-4 for high-res images, 5-7 for lower resolution

---

Scale Multiplication factor (Scale Multiplier)

• Default: 2.1
• Range: 1.5-3.0
• Description: Multiplication factor between successive scale levels. Controls how feature sizes progress from fine to coarse.
• Guidelines:
  • Lower values (1.5-2.0): Smaller steps between scales, more thorough coverage, slower processing
  • Higher values (2.5-3.0): Larger jumps between scales, faster but may miss intermediate-sized features
• Recommendation: 2.1 provides good coverage; use 1.7 for dense sampling, 2.5 for speed

---

Sigma/F (Frequency Bandwidth)

• Default: 0.65
• Range: 0.4-1.0
• Description: Bandwidth of the log-Gabor filters in frequency domain. Controls frequency selectivity.
• Guidelines:
  • Lower values (0.4-0.55): Narrow bandwidth, more selective to specific frequencies, sharper features but may miss some edges
  • Higher values (0.7-1.0): Broader bandwidth, detects wider range of frequencies, more complete edge detection but less selective
• Recommendation: 0.55-0.65 for sharp edges, 0.7-0.85 for textured or natural images

---

Angular spread (Angular Tolerance)

• Default: 15.0°
• Range: 5.0°-45.0°
• Description: Angular spread of each orientation filter in degrees. Controls directional selectivity.
• Guidelines:
  • Lower values (5-10°): Tight angular selectivity, good for detecting specific orientations, may fragment curved edges
  • Higher values (20-45°): Broader angular coverage, better edge continuity, less orientation-specific
• Recommendation: 15° for general use, 5-10° for straight edges, 20-30° for curved features

---

fd_nms_thresh_ (Additional Threshold)

• Default: 0.0 (disabled)
• Range: 0.0-0.3
• Description: Extra threshold applied after automatic (Otsu) thresholding. Values are normalized (0-1).
• Guidelines:
  • 0.0: Uses only automatic thresholding
  • 0.05-0.1: Mild suppression of weak responses, reduces noise
  • 0.15-0.3: Aggressive suppression, keeps only strongest features
• Recommendation: Leave at 0 for automatic operation, use 0.05-0.1 to clean up noisy results

---

fd_auto_run_ (Auto-Execute)

• Default: false
• Description: Controls whether detection runs automatically when parameters change.
• Guidelines:
  • false: Manual execution required (better for parameter tuning)
  • true: Immediate feedback on parameter changes (useful for interactive adjustment)
• Recommendation: Enable during parameter exploration, disable for batch processing

---

Quick Setup Presets

Clear, High-Contrast Images

Best for technical drawings, CAD images, or high-quality scans with sharp edges.

Parameter	Value
fd_scales_	3
fd_orientations_	8
fd_min_wavelength_	3.0
fd_mult_	2.5
fd_sigma_on_f_	0.55
fd_orient_bandwidth_deg_	10.0
fd_nms_thresh_	0.0

Natural/Textured Images

Optimal for photographs, geological images, or surfaces with natural textures.

Parameter	Value
fd_scales_	5
fd_orientations_	12
fd_min_wavelength_	4.0
fd_mult_	2.1
fd_sigma_on_f_	0.75
fd_orient_bandwidth_deg_	20.0
fd_nms_thresh_	0.05

Fast Processing

Balanced settings for real-time or near real-time processing needs.

Parameter	Value
fd_scales_	3
fd_orientations_	6
fd_min_wavelength_	5.0
fd_mult_	2.5
fd_sigma_on_f_	0.65
fd_orient_bandwidth_deg_	15.0
fd_nms_thresh_	0.0

Maximum Quality

Best possible detection quality when processing time is not a concern.

Parameter	Value
fd_scales_	6
fd_orientations_	16
fd_min_wavelength_	3.0
fd_mult_	1.8
fd_sigma_on_f_	0.65
fd_orient_bandwidth_deg_	15.0
fd_nms_thresh_	0.0

---

Parameter Interaction Guide

For Straight Linear Features

• Decrease fd_orient_bandwidth_deg_ (5-10°)
• Decrease fd_orientations_ (6-8)
• Decrease fd_sigma_on_f_ (0.45-0.55)

For Curved or Complex Features

• Increase fd_orient_bandwidth_deg_ (20-30°)
• Increase fd_orientations_ (12-16)
• Increase fd_scales_ (5-6)

To Reduce Noise

• Increase fd_min_wavelength_ (5-7)
• Increase fd_nms_thresh_ (0.05-0.15)
• Decrease fd_sigma_on_f_ (0.45-0.55)

To Detect Finer Details

• Decrease fd_min_wavelength_ (2-3)
• Increase fd_scales_ (5-7)
• Decrease fd_mult_ (1.5-1.8)

For Faster Processing

• Decrease fd_scales_ (2-3)
• Decrease fd_orientations_ (4-6)
• Increase fd_min_wavelength_ (5-7)
• Increase fd_mult_ (2.5-3.0)

---

Understanding the Algorithm

What is Phase Congruence?

Phase congruence is a feature detection method that identifies points where Fourier components are maximally in phase. Unlike gradient-based edge detectors, it:

• Is invariant to illumination and contrast changes
• Detects features independent of their magnitude
• Provides a dimensionless measure of feature significance

Multi-Scale Processing

The detector applies log-Gabor filters at multiple scales (controlled by fd_scales_ and fd_mult_) to capture features of different sizes in a single pass.

Orientation Selectivity

By using oriented filters (controlled by fd_orientations_ and fd_orient_bandwidth_deg_), the detector can identify edges at specific angles or across all orientations.

Advantages Over Traditional Edge Detection

• Illumination invariant: Works equally well in bright and dark regions
• Contrast independent: Detects both high and low contrast edges
• Multi-scale: Captures features at multiple scales simultaneously
• Phase-based: More robust to noise than gradient-based methods

---

Troubleshooting

Problem: Too Many False Detections

• Solution: Increase fd_nms_thresh_ to 0.05-0.1
• Alternative: Increase fd_min_wavelength_ to ignore fine noise

Problem: Missing Weak Edges

• Solution: Set fd_nms_thresh_ to 0.0
• Alternative: Increase fd_sigma_on_f_ for broader frequency response

Problem: Fragmented Edges

• Solution: Increase fd_orient_bandwidth_deg_ to 20-30°
• Alternative: Decrease fd_orientations_ for coarser angular sampling

Problem: Processing Too Slow

• Solution: Reduce fd_scales_ to 2-3 and fd_orientations_ to 4-6
• Alternative: Increase fd_min_wavelength_ and fd_mult_

Problem: Missing Fine Details

• Solution: Decrease fd_min_wavelength_ to 2-3
• Alternative: Increase fd_scales_ for better multi-scale coverage