Classification with Scattering Operators

TL;DR

This paper introduces scattering operators as local descriptors that leverage wavelet transforms to achieve translation invariance and deformation linearization, improving classification tasks like digit and texture recognition.

Contribution

It presents a novel scattering representation using wavelet cascades and demonstrates its effectiveness with a PCA-based supervised classification method.

Findings

Achieves state-of-the-art results in handwritten digit recognition.

Effective in texture classification tasks.

Provides a translation-invariant and deformation-stable feature representation.

Abstract

A scattering vector is a local descriptor including multiscale and multi-direction co-occurrence information. It is computed with a cascade of wavelet decompositions and complex modulus. This scattering representation is locally translation invariant and linearizes deformations. A supervised classification algorithm is computed with a PCA model selection on scattering vectors. State of the art results are obtained for handwritten digit recognition and texture classification.