Multiple-Kernel Local-Patch Descriptor
This addresses patch matching accuracy for computer vision applications, but it is incremental as it builds on existing handcrafted descriptor methods.
The paper tackled the problem of local-patch descriptor robustness to miss-registration by proposing a multiple-kernel descriptor based on gradient parametrizations, resulting in consistent outperformance of state-of-the-art methods on two benchmarks.
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.