Classification of curves in 2D and 3D via affine integral signatures (Q973791): Difference between revisions

In this paper special integral variables are assigned to a curve \(\gamma(t) = (x_1(t), \dots x_n(t))\) of \({\mathbb R}^n\). The integral variables are defined via \(x^{[\alpha_1, \dots,\alpha_n]}_i := \int_0^t x_1^{\alpha_1}(t) \dots x_n^{\alpha_n}(t) dx_i(t)\) with non-negative integers \(\alpha_1, \dots, \alpha_n\) satisfying \(\alpha_1 + \dots + \alpha_{i-1} + \alpha_{i+1} + \dots + \alpha_n\not = 0\). Based on these integral variables the authors derive two global invariants \(\tilde{I}_1 \tilde{I}_2\), with respect to the affine group for each planar curve (\(n=2\)). Similarly, a pair \(\tilde{J}_1, \tilde{J}_2\) of global affine invariants is derived in case of a spatial curve (\(n=3\)). In this way a 2-dimensional global \textit{signature curve} \((\tilde{I}_1(t), \tilde{I}_2(t))\) or \((\tilde{J}_1(t), \tilde{J}_2(t))\) can be assigned to each planar or spatial curve. Curves which are equivalent with respect to the affine group have the same signature curve. The attached signature curve is independent of the parametrization of the curve \(\gamma(t)\) but depends on the choice of an initial point on the curve, and hence cannot be used for local comparison. For that reason the authors additionally define a local signature curve for each planar or spatial curve. This local signature is again based on the integral variables but has the advantage of being independent of \textit{both} the choice of the initial point \textit{and} the parametrization. The main benefit of the invariants introduced in the paper is their robustness with respect to noise, a property which the classical differential invariants of curves miss due to fact that their definition is based on (higher order) derivatives.