Speeded Up Robust Features
Speeded Up Robust Features (SURF) was proposed by Herbert Bay, Tinne Tuytelaars, and Luc Van Gool in 2006. Some of the drawbacks of SIFT are that it is slow and computationally expensive. To target this problem, SURF was thought of. Apart from the increase in speed, the other motivations behind SURF were as follows:
Fast interest point detection
Distinctive interest point description
Speeded up descriptor matching
Invariant to the following common image transformations:
Image rotation
Scale changes
Illumination changes
Small changes in viewpoint
SURF detector
Just as SIFT approximate Laplacian of Gaussian images to Difference of Gaussian, SURF uses integral images to approximate Laplacian of Gaussian images. An integral image (summed area tables) is an intermediate representation of the image and contains the sum of grayscale pixel values of the image. It is called the fast Hessian detector. The descriptor, on the other hand, describes a distribution of Haar wavelet responses...
