Page 1
Lecture 3.2.0Introduction to keypoint features
Trym Vegard Haavardsholm
Page 2
Why extract features?
2
• Example: Panorama stitching– How do we combine two images?
Page 3
Why extract features?
3
• Example: Panorama stitching– How do we combine two images?
Step 1: Extract features
Page 4
Why extract features?
4
• Example: Panorama stitching– How do we combine two images?
Step 1: Extract featuresStep 2: Match features
Page 5
Why extract features?
5
• Example: Panorama stitching– How do we combine two images?
Step 1: Extract featuresStep 2: Match featuresStep 3: Align images
Page 8
Correspondences across views
8
Page 9
Correspondences across views
9
Page 10
Correspondences across views
10
Page 11
Correspondences across views
11
Page 12
Characteristics of good features
• Repeatability
12
Page 13
Characteristics of good features
• Repeatability• Distinctiveness
13
Page 14
Characteristics of good features
• Repeatability• Distinctiveness
• Efficiency
14
Page 15
Characteristics of good features
• Repeatability• Distinctiveness
• Efficiency• Locality
15
Page 16
Applications: Automatic Panoramas
16
Credit: Matt Brown
Page 17
Application: Object Recognition
17
Page 18
Application: 3D reconstruction and navigation
18
Page 19
Use of features in this course
• This week:– Lecture 3.2.1: Corner features– Lecture 3.2.2: Blob features
• Next week:– Feature descriptors– Feature matching– Homography estimation from correspondences
• Weeks after that:– Part II: World geometry and 3D– Part III: Scene analysis
19