Random projections for feature matching
Brendan McCane · 2014
Several methods of random projection for feature matching are investigated. These include: random projections of established feature descriptors for dimensionality reduction; random projections of raw image data as feature descriptors; random projections of raw image data as keypoint detectors; and random projections as stereo keypoint detectors and descriptors. Results indicate that random projections as stereo keypoint detectors produce accurate correspondences, but are computationally costly; and that random projections of established descriptors are both efficient and accurate.