HarrisZ+: Harris corner selection for next-gen image matching pipelines

Due to its role in many computer vision tasks, image matching has been subjected an active investigation by researchers, which lead better and more discriminant feature descriptors robust strategies, also thanks the advent of deep learning increased computational power modern hardware. Despite these achievements, keypoint extraction process at base pipeline not seen equivalent progresses. This paper presents HarrisZ+, upgrade HarrisZ corner detector, optimized synergically take advance recent improvements other steps pipeline. HarrisZ+ does only consists a tuning setup parameters, but introduces further refinements selection criteria delineated HarrisZ, so providing more, yet discriminative, keypoints, are distributed on with higher localization accuracy. The including together components, obtained different benchmarks state-of-the-art results among classic pipelines. These quite close those fully end-to-end trainable approaches show that there is still proper margin improvement can be granted research methods.