Distinctive Image Features from Scale-Invariant Keypoints

Distinctive Image Features from Scale-Invariant Keypoints
David G. Lowe
Computer Science Department
University of British Columbia
Vancouver, B.C., Canada
lowe@cs.ubc.ca
January 5, 2004
===============================================

This paper presents a method for extracting distinctive invariant features from
images that can be used to perform reliable matching between different views of
an object or scene.

There foir main steps for the feature extraction

1. Scale-space extrema detection:
It is implemented efficiently by using a difference-of-Gaussian function to identify potential interest points that are invariant to scale and orientation
Thu formula is as below,

the picture below shows the idea of the difference-of-Gaussian

2. Keypoint localization:
At each candidate location, a detailed model is fit to determine location and scale. Keypoints are selected based on measures of their stability.Some interest points might be not stable.  Aim to delete them from our candidate list.

3. Orientation assignment:
One or more orientations are assigned to each keypoint location
Based on local image gradient directions. All future operations are performed on image data that has been transformed relative to the assigned orientation, scale, and location for each feature, thereby providing invariance to these transformations. The way of caculating the  orientation is as follow,

orientation histogram uses 36 bins to quantize 360°

4. Keypoint descriptor:
The local image gradients are measured at the selected scale in the region around each keypoint. After
calculate the orientation and magnitude, and then accumulated into orientation histograms summarizing the contents over 4x4 subregions.Like the figure shows below.

These are transformed into a representation that allows for significant levels of local shape distortion and change in illumination.
====================================================
Comment

SIFT is really a very power and state-of-art descriptor to handle scaling and rotation problems. It 's a local feature that has the advantages which global features like color historgram don't have. While SIFT doesn't contain any information about color through the whole paper.