Gauge-SURF descriptors

Abstract

In this paper, we present a novel family of multiscale local feature descriptors, a theoretically and intuitively well justified variant of SURF which is straightforward to implement but which nevertheless is capable of demonstrably better performance with comparable computational cost. Our family of descriptors, called Gauge-SURF (G-SURF), is based on second-order multiscale gauge derivatives. While the standard derivatives used to build a SURF descriptor are all relative to a single chosen orientation, gauge derivatives are evaluated relative to the gradient direction at every pixel. Like standard SURF descriptors, G-SURF descriptors are fast to compute due to the use of integral images, but have extra matching robustness due to the extra invariance offered by gauge derivatives. We present extensive experimental image matching results on the Mikolajczyk and Schmid dataset which show the clear advantages of our family of descriptors against first-order local derivatives based descriptors such as: SURF, Modified-SURF (M-SURF) and SIFT, in both standard and upright forms. In addition, we also show experimental results on large-scale 3D Structure from Motion (SfM) and visual categorization applications.