A novel set of rotationally and translationally invariant features for images based on the non-commutative bispectrum

Abstract

We propose a new set of rotationally and translationally invariant features for image or pattern recognition and classification. The new features are cubic polynomials in the pixel intensities and provide a richer representation of the original image than most existing systems of invariants. Our construction is based on the generalization of the concept of bispectrum to the three-dimensional rotation group SO(3), and a projection of the image onto the sphere.

The claim that the invariants uniquely determine the original image had to be dropped

Cited by in corpus (10)

• On representing chemical environments
• Group Equivariant Convolutional Networks
• Bispectrum Inversion with Application to Multireference Alignment
• Machine learning nonequilibrium electron forces for adiabatic spin dynamics
• Anomalous phase separation dynamics in a correlated electron system: machine-learning enabled large-scale kinetic Monte Carlo simulations
• Image recovery from rotational and translational invariants
• Smart energy models for atomistic simulations using a DFT-driven multifidelity approach
• Discriminative Learning of Similarity and Group Equivariant Representations
• Unsupervised Co-Learning on $\mathcal{G}$-Manifolds Across Irreducible Representations
• Implicit Bias of Linear Equivariant Networks