Performance of Interpolated Histogram of Oriented Gradients on the Feature Calculation of SIFT

Abstract

Scale Invariant Feature Transform (SIFT) is the most dominant and robust object detection algorithm. It utilizes the Histogram of Oriented Gradients (HOG) method for feature computation. HOG is applied with trilinear interpolation to gain performance improvement. This paper examines the effect of interpolation on the performance of SIFT on both OXFORD and HPatches datasets. The various algorithms of interpolation for HOG, and the spatial binning process algorithm, are presented here. The performance is evaluated with Intersection Over Union, Correct Match Percentage, as well as the execution time of the algorithms. Moreover, we used the multiplication of the Intersection Over Union and Correct Match Percentage to take advantage of both metrics. It was observed that the interpolation did not significantly affect the performance of the SIFT.