Viirya

"Scale & Affine Invariant Interest Point Detectors"

By KRYSTIAN MIKOLAJCZYK AND CORDELIA SCHMID

--

In this paper, the authors propose a novel approach to detect interest points invariant to scale and affine transformations.

Their scale and affine invariant detectors are based on three results:

1. Interest points detected by Harris detector can be adapted to affine transformation.
2. The characteristic scale of a local structure is indicated by a local extrema over.
3. The affine shape of a point neighborhood is estimated based on second moment matrix.

Although Harris detector compared to other existing approaches demonstrates an excellent performance in previous review, it is not invariant to scale changes. This paper proposes a new interest point detector that combines Harris detector and automatic scale selection for a scale invariant detector.

This detector, which is called Harris-Laplace Detector, uses the scale-adapted Harris function to localize points in scale-space. Then it selects the points for which the Laplacian-of-Gaussian attains a maximum over scale.

Since Harris-Laplace Detector is not affine invariant, the authors extend it to the Harris-Affine Interest Point Detector.

In order to estimate the affine shape using second moment matrix, this approach automatically selects integration and differentiation scales.