Image Enhancement in Spatial Domain: Techniques, Filters & Applications

Core Concepts

1. Spatial Domain vs. Frequency Domain

• Spatial domain: Direct manipulation of pixel intensities in the image (e.g., brightness, contrast).
• Frequency domain: Manipulation of image frequencies (Fourier transform) to enhance edges or remove noise.
• Key difference: Spatial methods work on pixel values; frequency methods work on transformed data.

---

2. Point Operations (Intensity Transformations)

Definitions & Types

• Point operation: Modifies pixel values based on its own intensity (no neighbor dependency).
• Common operations:
  • Linear: $s=T(r)=a\cdot r+b$ (e.g., contrast stretching).
  • Nonlinear: Logarithmic, power-law (gamma correction), histogram equalization.

How It Works

• Input: Original image $f(x,y)$ with intensity $r$.
• Output: Transformed image $g(x,y)=T(r)$.
• Example: Dark image → contrast stretching to use full dynamic range.

Worked Example: Contrast St stretching

Given:

• Original image: 8-bit grayscale (0–255).
• Input range: $r_{\text{min}}=50$, $r_{\text{max}}=180$.
• Output range: 0–255.

Formula: $$s=\frac{(r-r_{\text{min}})}{(r_{\text{max}}-r_{\text{min}})}\cdot (L-1)$$ where $L=256$ (for 8-bit).

Steps:

1. For $r=50$: $s=0$.
2. For $r=180$: $s=255$.
3. For $r=100$: $$s=\frac{(100-50)}{(180-50)}\cdot 255=\frac{50}{130}\cdot 255\approx 98$$

Visualization:

Advantages:

• Simple, fast (no neighbor computation).
• Preserves spatial relationships.

Disadvantages:

• Limited to global changes (no local adaptability).

---

3. Spatial Filtering (Neighborhood Operations)

Definitions

• Spatial filter: Modifies pixel values based on a kernel/mask (e.g., 3×3, 5×5).
• Types:
  • Linear filters: Weighted average (e.g., mean, Gaussian blur).
  • Nonlinear filters: Median filter (noise removal), max/min filters.

How It Works

1. Kernel placement: Center the kernel over each pixel.
2. Convolution: Multiply kernel weights by pixel values, sum results.
3. Normalization: Divide by sum of kernel weights (for linear filters).

Worked Example: Mean Filter (Blurring)

Given:

• Image pixel values (3×3 neighborhood):

  [100 120 150]
  [110 130 140]
  [160 170 180]
  

• Kernel (3×3 mean filter): $$\frac{1}{9}\left[\begin{array}{ccc}1& 1& 1\\ 1& 1& 1\\ 1& 1& 1\end{array}\right]$$

Calculation for center pixel (130): $$\text{New value}=\frac{100+120+150+110+130+140+160+170+180}{9}=\frac{1260}{9}=140$$

Visualization:

Advantages:

• Reduces noise (smoothing).
• Preserves edges poorly (blurring effect).

Disadvantages:

• Blurs edges and fine details.
• Slow for large kernels.

---

4. Morphological Operations

Definitions

• Morphology: Shapes image based on structuring elements (e.g., disk, square).
• Operations:
  • Erosion: Shrinks bright regions (removes small white noise).
  • Dilation: Expands bright regions (fills gaps).
  • Opening: Erosion → Dilation (removes small objects).
  • Closing: Dilation → Erosion (fills small holes).

How It Works

1. Erosion: Replace pixel with min of neighborhood (under structuring element).
2. Dilation: Replace pixel with max of neighborhood.

Worked Example: Erosion with Disk Structuring Element

Given:

• Binary image (1 = object, 0 = background):

  [1 1 0]
  [1 0 0]
  [0 0 0]
  

• Structuring element (3×3 disk):

  [0 1 0]
  [1 1 1]
  [0 1 0]
  

Erosion of center pixel (0):

• Overlay structuring element → all 8 neighbors must be 1 (not true here).
• Result: Center becomes 0 (unchanged, but edges shrink).

Visualization:

Advantages:

• Preserves shape while removing noise.
• Useful for segmentation.

Disadvantages:

• Over-erosion can disconnect objects.
• Sensitive to structuring element size.

---

5. Applications in Real World

1. Medical Imaging (e.g., NAMS Hospital, Nepal)

• Use case: Enhancing X-ray images for better diagnosis.
• Technique: Histogram equalization to improve contrast in low-light X-rays.
• Impact: Helps radiologists detect fractures or tumors earlier.

2. Surveillance Systems (e.g., NTC Traffic Cameras)

• Use case: License plate recognition in poor lighting.
• Technique: Adaptive histogram equalization (AHE) to enhance plate visibility.
• Impact: Reduces false negatives in automated toll systems.

3. Agriculture (e.g., Daraz Farm Products)

• Use case: Detecting pests in crop images.
• Technique: Median filtering to remove noise from drone-captured images.
• Impact: Early pest detection → higher yield.

4. Social Media (e.g., Facebook Filters)

• Use case: Beauty filters (e.g., "Smooth Skin").
• Technique: Gaussian blur + edge-preserving filters.
• Impact: Millions of daily users rely on spatial filtering for aesthetic enhancement.

---

6. Comparison Table: Enhancement Techniques

---

7. Exam Tip

What Examiners Look For

1. Definitions: Clearly distinguish between point operations, spatial filtering, and morphology.
2. Mathematical Steps: Show full calculations for kernel applications (e.g., mean filter).
3. Visual Proof: Always draw before/after for contrast stretching or filtering.
4. Applications: Link techniques to real-world problems (e.g., medical imaging, surveillance).
5. Trade-offs: Discuss when to use linear vs. nonlinear filters (e.g., median for impulse noise).

Common Pitfalls

• Assuming all filters are interchangeable: Mean filter ≠ median filter!
• Ignoring kernel size: Larger kernels blur more but take longer.
• Forgetting normalization: Always divide by kernel sum for linear filters.

Sample Exam Question & Answer

Q: Explain how a 3×3 mean filter works. Apply it to a 3×3 neighborhood with pixels:

[50 60 70]
[80 90 100]
[110 120 130]

A:

1. Kernel: All weights = $\frac{1}{9}$.
2. Calculation for center (90): $$\text{New value}=\frac{50+60+70+80+90+100+110+120+130}{9}=\frac{810}{9}=90$$ (Note: Center remains unchanged in this symmetric case.)
3. Edge pixels: Use partial kernel (e.g., top-left pixel = average of its 3 neighbors).

---