Filtered projections as an aid in corner detection

Corners are very useful features for such purposes as image matching or shape analysis, but corner detection is a relatively expensive operation. This paper uses filtered x and y projections, applied to an image containing an object that has not been explicitly segmented from its background, to determine possible positions of corners, so that corner detection can be applied only in the vicinity of these positions. Even in cases where the object would be hard to segment (unimodal histogram), this approach yields a good set of possible corner positions.