Foundations

Segmentation & edges

Algorithm

Intermediate

Laplacian and Laplacian of Gaussian

Use second-derivative style filters to emphasize rapid intensity change and zero crossings, often after smoothing first.

Segmentation & edges is nested under Foundations , so the broader pipeline usually still applies here.

graphics

image-processing

edges

second-derivative

Interactive playground

Tweak the operator or scene live so the article connects to an immediate visual result.

Laplacian / LoG

Smooth first, then look for rapid second-derivative changes that highlight ridges, blobs, and edge-like structure.

Input image

LoG response magnitude

96 × 96 live preview

The radius controls the smoothing scale before the Laplacian reacts to local curvature.

Kernel radius 2

Response strength 1.6×

Family

Foundations -> Segmentation & edges

Turn raw pixels into masks, boundaries, and simpler structural signals.

Builds on

2 topics

Read these first if you want the surrounding pipeline context.

Unlocks

1 next topic

Use these follow-ups when you want to keep turning the image-processing pipeline forward.

Learning paths

This topic appears in curated graphics progressions so the next step is obvious.

Choose this over that

Second-derivative edges vs gradient edges

Laplacian-style operators react to zero crossings and rapid curvature, while Sobel and Canny focus on directed gradients.

Current topic

Laplacian and Laplacian of Gaussian

Choose this when: You want second-derivative style edge emphasis or blob-like structures after smoothing.

Choose something else when: You need a more stable thin edge map with directional gradient reasoning.

Segmentation & edges Open topic

Sobel Edge Detection

Choose this when: A simple gradient magnitude map is enough.

Choose something else when: You need multi-stage suppression and thresholding instead of a raw edge response.

Segmentation & edges Open topic

Canny Edge Detection

Choose this when: The result should be a cleaner final edge map rather than an intermediate signal.

Choose something else when: A lighter-weight operator already answers the need.

Problem

Gradient filters tell you how strongly intensity changes in a direction, but sometimes you care about curvature and zero crossings instead: where the signal bends through a sharp transition after noise has been tamed.

Intuition

The Laplacian is a second derivative. It becomes large when local intensity bends rapidly. Because second derivatives amplify noise, a common practical version is LoG: blur first, then apply the Laplacian.

Core idea

Apply a Laplacian kernel such as a 3 x 3 stencil with a strong negative center and positive neighbors.
For the LoG variant, smooth the image first with a Gaussian kernel before applying the second-derivative pass.
Inspect the response directly or look for zero crossings around strong responses.

Worked example

On a blurry bright blob against a dark background, the Laplacian becomes positive on one side of the transition and negative on the other. That sign change is often the useful signal rather than the raw magnitude alone.

Complexity

A small fixed Laplacian kernel is O(WH). LoG adds the blur step first, so the full pipeline cost is the blur plus one Laplacian pass.

When to choose it

Choose it when second-derivative behavior or blob-style structure matters.
Choose Sobel when a simpler gradient magnitude map is sufficient.
Choose Canny when the final output should be a carefully thinned edge map rather than an intermediate response.

Key takeaways

The Laplacian is a second-derivative operator.
It is noise-sensitive, so smoothing first is often the practical move.
LoG is best understood as blur plus curvature-sensitive edge emphasis.
It complements, rather than replaces, first-derivative operators such as Sobel.

Practice ideas

Run Sobel and Laplacian on the same image and compare what each one highlights.
Toggle a Gaussian pre-blur and inspect how noisy the raw Laplacian becomes without it.
Look for zero crossings instead of only bright magnitude responses.

Relation to other topics

Gaussian blur is the usual stabilizing pre-step for LoG.
Sobel measures first-derivative gradients rather than second-derivative curvature.
Canny turns gradient logic into a multi-stage edge pipeline with suppression and hysteresis.

Build on these first

These topics supply the mental model or preceding stage that this page assumes.

Foundations

Concept

Image Processing Fundamentals

Build the mental model behind blur, thresholding, edges, and morphology: pixels, neighborhoods, kernels, masks, and why local operators compose into pipelines.

Filtering & convolution

Algorithm