Rigid-Motion Scattering for Texture Classification

TL;DR

This paper introduces a rigid-motion scattering method that computes rotation and translation invariant features for texture classification, achieving state-of-the-art results on diverse datasets with rotation and scale variations.

Contribution

It develops a deep convolutional framework on the rigid-motion group using wavelets, preserving joint rotation and translation info for improved texture analysis.

Findings

Achieved state-of-the-art accuracy on multiple texture datasets.

Effectively handles rotation and scaling variabilities.

Provides a mathematically grounded invariant feature extraction method.

Abstract

A rigid-motion scattering computes adaptive invariants along translations and rotations, with a deep convolutional network. Convolutions are calculated on the rigid-motion group, with wavelets defined on the translation and rotation variables. It preserves joint rotation and translation information, while providing global invariants at any desired scale. Texture classification is studied, through the characterization of stationary processes from a single realization. State-of-the-art results are obtained on multiple texture data bases, with important rotation and scaling variabilities.