Multiple-Kernel Local-Patch Descriptor

TL;DR

This paper introduces a novel local-patch descriptor that combines multiple kernels to improve robustness against patch misregistration, outperforming existing methods on benchmark tests.

Contribution

It presents a new multiple-kernel descriptor based on efficient match kernels of patch gradients, integrating two parametrizations for enhanced robustness.

Findings

Outperforms state-of-the-art methods on local patch benchmarks

Combines polar and Cartesian parametrizations for robustness

Effective against patch mis-registration and orientation noise

Abstract

We propose a multiple-kernel local-patch descriptor based on efficient match kernels of patch gradients. It combines two parametrizations of gradient position and direction, each parametrization provides robustness to a different type of patch miss-registration: polar parametrization for noise in the patch dominant orientation detection, Cartesian for imprecise location of the feature point. Even though handcrafted, the proposed method consistently outperforms the state-of-the-art methods on two local patch benchmarks.