Scale Invariant Feature Transform

Scale Invariant Feature Transform (SIFT) is one of the most widely recognized feature detection algorithms. It was proposed by David Lowe in 2004.

Note

Link to the paper: http://www.cs.ubc.ca/~lowe/papers/ijcv04.pdf

Some of the properties of SIFT are as follows:

• It is invariant to scaling and rotation changes in objects

• It is also partially invariant to 3D viewpoint and illumination changes

• A large number of keypoints (features) can be extracted from a single image

Understanding how SIFT works

SIFT follows a strategy of matching robust local features. It is divided into four parts:

• Scale-space extrema detection

• Keypoint localization

• Orientation assignment

• Keypoint descriptor

Scale-space extrema detection

In this step, an image is progressively blurred out using Gaussian blur to get rid of some details in the images. It has been mathematically proven (under reasonable assumptions) that performing Gaussian blur is the only way to carry this out effectively.

Images of one...