CoreSource checked

SIFT

Detects and describes scale-invariant image keypoints from local gradient structure.

Explore relationships Open constellationBrowse atlas

Quick Facts

Compact metadata for implementation and discovery.

Category: Image Processing
Difficulty: Intermediate
Family: Feature Detection
Tier: Core
Status: Verified
Review: Source Checked
Sources: 2
Formulas: 0
Relationships: 7
Code example: Ready
Visualization: Ready
Aliases: Scale-Invariant Feature Transform

Overview

SIFT finds candidate keypoints across scale space, refines their locations, assigns orientations, and builds local gradient histograms that are robust to moderate scale and rotation changes.

Core Concepts

scale-invariant feature detectionfeature detectionfeature matchingimage processing

Quick facts

Complexity

Runtime and memory behavior at a glance.

Best

O(n)

Average

O(n log n)

Worst

O(n^2)

Space

O(n)

When to Use

Use when repeatable local features are needed under moderate scale and rotation changes.
Use when descriptor quality is more important than minimal computation.

When Not to Use

Avoid when binary descriptors are needed for very fast matching.
Avoid when severe viewpoint, blur, or textureless regions dominate the image.

What It Does

Builds a scale-space representation
Detects stable extrema as keypoints
Computes orientation-normalized local descriptors

Use Cases

Image matching and registration
Object recognition with classical features
Recovering correspondences before homography or 3D reconstruction

Inputs

2D image intensity array I(x, y)
Scale-space octave parameters and contrast/corner thresholds

Outputs

Scale- and rotation-invariant keypoints (x, y, σ, θ)
Local descriptor vectors (floating-point or binary)

Pseudocode

1.Build Gaussian and difference-of-Gaussian scale spaces
2.Find local extrema across position and scale
3.Reject unstable low-contrast or edge-like keypoints
4.Assign dominant orientations from local gradients
5.Build and normalize the keypoint descriptor

Canonical Example

Input: two images of the same planar object at different scales
Output: matched SIFT keypoints between the images
Explanation: Stable scale-space keypoints receive descriptors that can be matched despite scale changes.

Implementation Notes

Keep octave, scale, contrast, and edge-response thresholds explicit.
Use descriptor normalization and clipping consistently.
Validate matches with a geometric model such as RANSAC homography when appropriate.

Common Pitfalls

Skipping keypoint refinement leaves unstable low-contrast or edge-like points.
Mismatched descriptor normalization changes nearest-neighbor behavior.
Using raw nearest matches without ratio tests or geometric verification creates false correspondences.