Exercise 1: The Scale-Invariant Feature
Transform
August 9, 2010
1 Introduction
The Scale-Invariant Feature Transform (SIFT) is a method for the detection and de-
scription of interest points developed by David Lowe [Lowe, 2004]. The method has
been successfully applied in many computer- and machine-vision applications, such as
object recognition, image retrieval, and robotic localization and mapping.
In this exercise, we will use SIFT for the representation of the environment of a robot.
We will use the detector to find interest points in the environment, and we will describe
those interest points with the SIFT descriptor, so that we will be able to store and
recognize the points. The main goal of this exercise is to play around with the different
parameters of the detector and descriptor to get a better grasp on the method.
2 Exercises
The detector is based on the filtering of the image with Difference of Gaussians:
D(σ) = G(σ) − G(kσ), a so-called Mexican-hat function. Lets visualize the Differ-
ence of Gaussian:
2.1 DIfference-of-Gaussian
• Start matlab
• Change directory to exercise_1.
• Create a Gaussian filter kernel of size 101 × 101 and σ
1
= 20.0 with
G1 = fspecial(’gaussian’, [101 101], 20.0);
Mind that the size of the kernel is much bigger than is actually used in the SIFT
method, but this is for displaying purposes.
1
评论0