KEMBAR78
SIFT Detector Descriptor | PDF | Algorithms | Vision
Scribd
Upload
14 views42 pages

SIFT Detector Descriptor

The document provides an overview of the Scale Invariant Feature Transform (SIFT) technique used in image recognition and alignment. It details the process of extracting SIFT interest points, computing principal orientations, and creating descriptors that are invariant to rotation, scale, and brightness. Additionally, it discusses applications such as panorama stitching and auto-collage creation using SIFT, while noting its limitations with significant viewpoint changes.

Uploaded by

Inas Tazarti
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
14 views42 pages

SIFT Detector Descriptor

The document provides an overview of the Scale Invariant Feature Transform (SIFT) technique used in image recognition and alignment. It details the process of extracting SIFT interest points, computing principal orientations, and creating descriptors that are invariant to rotation, scale, and brightness. Additionally, it discusses applications such as panorama stitching and auto-collage creation using SIFT, while noting its limitations with significant viewpoint changes.

Uploaded by

Inas Tazarti
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 42

SIFT

Computer Science Department


Ibn Zohr University
Master Data Science

Vision par Ordinateur


Overview
How do we recognize these objects?
Overview
How do we recognize these objects?
Overview
How do we recognize these objects?

Template 2D image

Find and match "Interesting points or features"


SIFT

Scale Invariant Feature Transform (SIFT) and its


use in image alignment and 2D image recognition.
Fast NLoG Approximation: DoG
Extracting SIFT Interest Points

Image
I(x, y)
Extracting SIFT Interest Points

Image
Image
I(x, y)
I(x, y)
Extracting SIFT Interest Points

Image
I(x, y)

Finding Extremum
in every 3x3x3 grid
Extracting SIFT Interest Points

Image
I(x, y)

Finding Extremum Interest Point Candidates


in every 3x3x3 grid
Extracting SIFT Interest Points

Image
I(x, y)

Interest Point Candidates


Extracting SIFT Interest Points

Image
I(x, y)

Interest Point Candidates SIFT Interest Points (after


removing weak extrema)
Extracting SIFT Interest Points

Image
I(x, y)

Interest Point Candidates SIFT Interest Points (after


removing weak extrema)
Extracting SIFT Interest Points

Image
I(x, y)

Interest Point Candidates SIFT Interest Points (after


removing weak extrema)
Extracting SIFT Interest Points

Image
I(x, y)

Interest Point Candidates SIFT Interest Points (after


removing weak extrema)
Extracting SIFT Interest Points

Image
I(x, y)

Interest Point Candidates SIFT Interest Points (after Interest Points


removing weak extrema) Visualization
SIFT Detection Examples
SIFT Detection Examples
SIFT Scale Invariance
SIFT Scale Invariance
SIFT Scale Invariance
SIFT Scale Invariance
Computing the Principal Orientation
Use the Histogram of Gradient Directions

Image gradient directions


Computing the Principal Orientation
Use the Histogram of Gradient Directions

Image gradient directions Choose the most prominent


direction
Example of Computing Orientation

30 35 36 39 45
35 40 42 46 47
37 46 50 55 50
43 52 56 58 60
51 60 65 68 70

Magnitude:

Orientation:
SIFT Rotation Invariance
SIFT Rotation Invariance
Choose the principal orientation to undo rotation
SIFT Descriptor
Histograms of gradient directions over spatial regions
SIFT Descriptor
Histograms of gradient directions over spatial regions
SIFT Descriptor
Histograms of gradient directions over spatial regions
SIFT Descriptor
Histograms of gradient directions over spatial regions

Normalized histograms: Invariant to rotation, scale and brightness


Comparing SIFT Descriptors
Comparing SIFT Descriptors
Comparing SIFT Descriptors

Larger the distance metric, better the match.


SIFT Results: Scale Invariance
SIFT Results: Rotation Invariance
SIFT Results: Robustness to Clutter
Panorama Stitching using SIFT

Image 1 Image 2

Match SIFT Interest Points


Panorama Stitching using SIFT

Wrap and combine images to create a larger image


Auto Collage using SIFT

Captured images
Auto Collage using SIFT
Collage
SIFT for 3D Objects?

No change in viewpoint 30 degrees change in viewpoint

SIFT is reliable for only small changes in viewpoint.

You might also like