Corner detection in natural images based on the 2-D Hilbert transform

Corners are special features in images, and are of great use in applications such as analysing the structure of motion or estimating the depth in stereo images. Most existing corner detectors are relatively expensive, because they require the realisation of filter banks or, at least, the exact determination of the object contours. The algorithm proposed is based on the 2-D Hilbert transform, which can be numerically evaluated by a separable 2-D digital filter. The implementation of this corner detector employs only one single digital filter and in a second stage, a peak detector to localise the extrema in the filter output. The efficacy of the detector has been demonstrated by two examples.