Rotation invariant CNN using scattering transform for image classification

Rosemberg Rodriguez Salas (1) , Eva Dokladalova (2, 1) , Petr Dokládal (3)

1 ESIEE Paris
2 LIGM - Laboratoire d'Informatique Gaspard-Monge
3 CMM - Centre de Morphologie Mathématique

Deep convolutional neural networks accuracy is heavily impacted by rotations of the input data. In this paper, we propose a convolutional predictor that is invariant to rotations in the input. This architecture is capable of predicting the angular orientation without angle-annotated data. Furthermore, the predictor maps continuously the random rotation of the input to a circular space of the prediction. For this purpose, we use the roto-translation properties existing in the Scattering Transform Networks with a series of 3D Convolutions. We validate the results by training with upright and randomly rotated samples. This allows further applications of this work on fields like automatic re-orientation of randomly oriented datasets.

Mots clés: image classification; convolu- tional neural network; Index Terms-Rotation; invariant; image classification.; convolutional neural network; covariant; Rotation

Domaines: Intelligence artificielle [cs.AI]; Vision par ordinateur et reconnaissance de formes [cs.CV]; Réseau de neurones [cs.NE]

Rosemberg Rodriguez Salas, Eva Dokladalova, Petr Dokládal. Rotation invariant CNN using scattering transform for image classification. IEEE International Conference on Image Processing (ICIP), Sep 2019, Taipei, Taiwan. ⟨hal-02008378v1⟩