Convolution for Sharping and Edge Detection

🧠 Common confusion
High-pass, Laplacian, Sobel, Prewitt, Roberts…
Are these sharpening filters or edge detectors?

The short answer: they are the same operations used for different purposes.

This post clarifies the conceptual boundary between sharpening and edge extraction, using kernels, equations, and interpretation — not just names.

---

🔍 Start with the Key Distinction

The difference is not the kernel.
The difference is how the response is used.

• Edge detection
  → output is the edges

• Sharpening
  → edge-like response is added back to the original image

Same math. Different intent.

---

🧱 High-Pass Filtering (The Root Concept)

Idea

• Low frequencies → smooth structure
• High frequencies → edges, transitions, fine detail

A high-pass filter suppresses low-frequency content and keeps rapid changes.

---

Simple High-Pass Kernel (3×3)

-1  -1  -1
-1   8  -1
-1  -1  -1

or

 0  -1   0
-1   4  -1
 0  -1   0

---

Interpretation

• If you use this output directly → edge / detail map
• If you add it to the original image → sharpening

---

✨ Image Sharpening (Unsharp View)

Sharpening is often expressed as:

\[I_{sharp} = I + \alpha \cdot (I * H)\]

Where:

• $I$ = original image
• $H$ = high-pass filter
• $\alpha$ = sharpening strength

This is why sharpening feels like “boosting edges”.

---

🧮 Laplacian Filter

Kernel

 0  -1   0
-1   4  -1
 0  -1   0

or

-1  -1  -1
-1   8  -1
-1  -1  -1

Meaning

• Second-order derivative
• Responds to rapid intensity changes in all directions
• Very sensitive to noise

📌 Laplacian by itself → edge map
📌 Laplacian added to image → sharpening

---

➡️ First-Order Gradient Filters

These approximate first derivatives.

They respond to directional change.

---

🧭 Roberts Operator

Kernels

 1   0
 0  -1

 0   1
-1   0

Characteristics

• Very small support
• Extremely sensitive to noise
• Historically important

Mostly used for edge detection, rarely for sharpening.

---

🧭 Prewitt Operator

Kernels (Horizontal / Vertical)

-1  0  1
-1  0  1
-1  0  1

 1   1   1
 0   0   0
-1  -1  -1

Characteristics

• Simple gradient estimate
• Uniform weighting
• Moderate noise sensitivity

---

🧭 Sobel Operator

Kernels (Horizontal / Vertical)

-1  0  1
-2  0  2
-1  0  1

 1   2   1
 0   0   0
-1  -2  -1

Why Sobel Is Popular

• Includes smoothing in one direction
• More robust to noise than Prewitt
• Good balance between edge clarity and stability

---

🧠 Gradient Magnitude

From Sobel / Prewitt:

\[|\nabla I| = \sqrt{G_x^2 + G_y^2}\]

This is a pure edge representation.

Used directly → edge extraction
Added back → sharpening

---

🎯 So… Sharpening or Edge Detection?

Filter	Edge Detection	Sharpening
High-pass	✅	✅
Laplacian	✅	✅
Roberts	✅	❌
Prewitt	✅	⚠️
Sobel	✅	⚠️

⚠️ = possible, but uncommon

---

💡 A Vision-Centric Takeaway

Edges are information. Sharpening is emphasis.

• Edge detection isolates structure
• Sharpening reinjects structure into appearance

The math does not change.
The interpretation does.

---

✨ Summary

• Sharpening and edge detection share the same operators
• Difference lies in usage, not definition
• High-pass & Laplacian are most common for sharpening
• Gradient operators are primarily for edge extraction
• Understanding intent prevents conceptual confusion

---

🌈 In image processing, the same math can answer very different questions.