CS201 Computer Vision Lect 08: SIFT Keypoint Detection John Magee - PowerPoint PPT Presentation

CS201 – Computer Vision Lect 08: SIFT Keypoint Detection John Magee 23 Septermber 2014 Slides courtesy of Diane H. Theriault

Question of the Day: • How do we find repeatable, stable, scale- invariant points in images?

SIFT • Find repeatable, scale-invariant points in images (today) • Compute something about them • Use the thing you computed to perform matching • “Distinctive Image Features from Scale -Invariant Keypoints ” by David Lowe • Patented!

How to find the same cat? • Imagine that we had a library of cats • How could we find another picture of the same cat in the library? • Look for the markings?

How to find the same cat? • Imagine that we had a library of cats • How could we find another picture of the same cat in the library? • Look for the markings? • Which markings?

How to find the same cat? • Imagine that we had a library of cats • How could we find another picture of the same cat in the library? • Look for the markings? • Which markings?

How to find the same cat? • Imagine that we had a library of cats • How could we find another picture of the same cat in the library? • Look for the markings? • Which markings? • Which cat would be easier to uniquely identify?

Keypoints • Idea: Want to find points that are easy to localize uniquely This one Not this one

Keypoints at Different Scales • Idea: Want to find scale-invariant points that are easy to localize uniquely

Gaussian

Scale Space • Image convolved with Gaussians of different widths

Scale Space Octaves • Every time the width of your Gaussian doubles, downsample the image

Derivative of a Gaussian

Second Derivative of a Gaussian

Laplacian of a Gaussian • Sum of spatial second derivatives

Keypoints with Image Filtering Image • Perform image filtering by convolving an image with a “filter”/”mask” / “kernel” to obtain a “result” / “response” • The value of the result will be positive in regions of the image that “look like” the filter • What would a “dot” filter look like? Filter

Scale-space extrema • “ Extrema ” = local minimum or maximum • Check 8 neighbors at a particular scale • Check neighbors at scales above and below

Scale-space Extrema

Difference of Gaussians • Approximation of the Laplacian of a Gaussian

Discussion Questions: • What does it mean for an image point to be repeatable? • What are the properties of repeatable points?

Keypoints with Image Filtering • What tool have we learned about for finding places in the image that “look like” corners or dots?