1 Introduction

2024.4.17

a core problem of computer vision: Detection and description of 2D feature points for image matching.

1. Filtering & Transformation → Edges → Feature points

   1. Also called interest points, key points, etc.
   2. Often described as ‘local’ features.

2. Human eye movements

3. Your eyes don’t see everything at once, but they jump around.

   1. You see only about 2 degrees with high resolution

4. What catches your interest?

1.1 Contents

1. Local invariant features
   1. Motivation
   2. Requirements, invariances
2. Keypoint localization
   1. Harris corner detector
3. Scale invariant region selection
   1. Automatic scale selection
   2. Difference-of-Gaussian (DoG) detector
4. Some background reading:
   1. Rick Szeliski, Chapter 4.1;
   2. (optional) K. Mikolajczyk, C. Schmid, A performance evaluation of local descriptors. In PAMI 27(10):1615-1630
   3. http://www.robots.ox.ac.uk/~vgg/research/affine/det_eval_files/mikolajczyk_

1.2 Motivation for using local features

1. Global representations have major limitations
2. Instead, describe and match only local regions

1.2.1 Increased robustness to

• Occlusions

• Articulation

• Intra-category variations

  

1.3 Fundamental to Applications

Feature points are used for:

• Image alignment