Fast and Accurate Gaussian Derivatives based on B-Splines

Gaussian derivatives are often used as differential operators to analyze the structure in images. In this paper, we will analyze the accuracy and computational cost of the most common implementations for differentiation and interpolation of Gaussian-blurred multi-dimensional data. We show that – for the computation of multiple Gaussian derivatives – the method based on B-splines obtains a higher accuracy than the truncated Gaussian at equal computational cost.