HarrisZ$^+$: Harris Corner Selection for Next-Gen Image Matching Pipelines

Due to its role in many computer vision tasks, image matching has beensubjected to an active investigation by researchers, which has lead to betterand more discriminant feature descriptors and to more robust matchingstrategies, also thanks to the advent of the deep learning and the increasedcomputational power of the modern hardware. Despite of these achievements, thekeypoint extraction process at the base of the image matching pipeline has notseen equivalent progresses. This paper presents HarrisZ$^+$, an upgrade to theHarrisZ corner detector, optimized to synergically take advance of the recentimprovements of the other steps of the image matching pipeline. HarrisZ$^+$does not only consists of a tuning of the setup parameters, but introducesfurther refinements to the selection criteria delineated by HarrisZ, soproviding more, yet discriminative, keypoints, which are better distributed onthe image and with higher localization accuracy. The image matching pipelineincluding HarrisZ$^+$, together with the other modern components, obtained indifferent recent matching benchmarks state-of-the-art results among the classicimage matching pipelines. These results are quite close to those obtained bythe more recent fully deep end-to-end trainable approaches and show that thereis still a proper margin of improvement that can be granted by the research inclassic image matching methods.