ExtendedSource checked

Difference of Gaussians

Highlights image structures by subtracting two Gaussian-blurred versions at different scales.

Explore relationships Open constellationBrowse atlas

Quick Facts

Compact metadata for implementation and discovery.

Category: Image Processing
Difficulty: Intermediate
Family: Scale-Space Filtering
Tier: Extended
Status: Verified
Review: Source Checked
Sources: 2
Formulas: 1
Relationships: 1
Code example: Ready
Visualization: Ready

Overview

Difference of Gaussians approximates scale-selective edge or blob detection by comparing a lightly blurred image with a more heavily blurred one.

Core Concepts

Gaussian scaleimage subtractionscale spaceblob response

Quick facts

Complexity

Runtime and memory behavior at a glance.

Best

O(pixels)

Average

O(pixels * kernel_size)

Worst

O(pixels * kernel_size)

Space

O(pixels)

When to Use

Use when features at a chosen scale should stand out from smoother background content.
Use when blob detection is a central requirement.

When Not to Use

Avoid when raw edges are needed without scale filtering.
Avoid when lighting, noise, or scale changes dominate the image signal.

What It Does

Blurs the image at two scales
Subtracts the blurred images
Highlights scale-dependent changes

Use Cases

Blob detection
Edge enhancement
Feature detection preprocessing

Inputs

Image
Two Gaussian blur scales

Outputs

Difference image

Key Formulas (1)

DoG response

DoG = GaussianBlur(image, sigma1) − GaussianBlur(image, sigma2)

DoG = GaussianBlur(image, sigma1) - GaussianBlur(image, sigma2)

Compares image structure visible at two blur scales.

Pseudocode

1.Blur the image with a small sigma
2.Blur the image with a larger sigma
3.Subtract the second blurred image from the first
4.Optionally threshold or normalize the result

Canonical Example

Input: image has small bright spot, sigma1 = 1, sigma2 = 3
Output: spot appears strongly in difference image
Explanation: The small structure survives light blur more than heavy blur.

Implementation Notes

Reuse separable Gaussian blur for both scales to keep the operation efficient.
Start with grayscale or single-channel tests before applying the method to full images.
Keep a tiny example with expected output beside the implementation.

Common Pitfalls

Using nearly identical blur scales produces a weak noisy response.
Padding, threshold, and color-space choices can change the result dramatically.
Not defining boundary, empty, or degenerate inputs before implementation.

Interactive Mechanism Walkthrough

Spatial Primitive & Sweep-Line Canvas

Show fine details suppressed differently at two blur scales, then emphasized by subtraction.

Step 1 of 4: Blur the image with a small sigma

image-kernel

Notes

The choice of sigma values controls which structures are emphasized.

Sources (2)

What each reference supports for this entry.

Computer Vision: Algorithms and Applications

Book

Computer-vision reference supporting Difference of Gaussians feature extraction, image registration, segmentation, or multiscale image-analysis workflow.

Open source