Chapter 6 - Corner, interest point, and invariant feature detection

Abstract

This chapter is concerned with the design of corner and interest point detectors. It starts by examining second-order derivative schemes for corner detection, and then shows that median filter–based detectors give very similar responses. Next, it goes on to consider the widely used Harris interest point operator, which is based on the determinant of the matrix of second moments of the first derivatives of the intensity function. Having dealt with basic schemes for corner and interest point detection, attention is focused on how to make detectors scale and affine invariant for applications such as wide baseline stereo. In this context the scale invariant feature transforms, speeded-up robust features, maximally stable extremal region, histograms of oriented gradient and other detectors, and the types of descriptor they produce are examined: performance evaluation criteria are also examined carefully, and particular reference is made to accuracy and repeatability.