On the Robustness of Convolutional Neural Networks Regarding Transformed Input Images

Frederik Timme, Jochen Kerdels, Gabriele Peters

2020

Abstract

Convolutional Neural Networks (CNNs) have become the dominant and arguably most successful approach for the task of image classification since the release of AlexNet in 2012. Despite their excellent performance, CNNs continue to suffer from a still poorly understood lack of robustness when confronted with adversarial attacks or particular forms of handcrafted datasets. Here we investigate how the recognition performance of three widely used CNN architectures (AlexNet, VGG19 and ResNeXt) changes in response to certain input data transformations. 10,000 images from the ILSVRC2012s validation dataset were systematically manipulated by means of common transformations (translation, rotation, color change, background replacement) as well as methods like image collages and jigsaw-like puzzles. Both the effect of single and combined transformations are investigated. Our results show that three of these input image manipulations (rotation, collage, and puzzle) can cause a significant drop in classification accuracy in all evaluated architectures. In general, the more recent VGG19 and ResNeXt displayed a higher robustness than AlexNet in our experiments indicating that some progress has been made to harden the CNN approach against malicious or unforeseen input.

Paper Citation

in Harvard Style

Timme F., Kerdels J. and Peters G. (2020). On the Robustness of Convolutional Neural Networks Regarding Transformed Input Images. In Proceedings of the 12th International Joint Conference on Computational Intelligence (IJCCI 2020) - Volume 1: NCTA; ISBN 978-989-758-475-6, SciTePress, pages 396-403. DOI: 10.5220/0010107203960403

in Bibtex Style

@conference{ncta20,
author={Frederik Timme and Jochen Kerdels and Gabriele Peters},
title={On the Robustness of Convolutional Neural Networks Regarding Transformed Input Images},
booktitle={Proceedings of the 12th International Joint Conference on Computational Intelligence (IJCCI 2020) - Volume 1: NCTA},
year={2020},
pages={396-403},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0010107203960403},
isbn={978-989-758-475-6},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 12th International Joint Conference on Computational Intelligence (IJCCI 2020) - Volume 1: NCTA
TI - On the Robustness of Convolutional Neural Networks Regarding Transformed Input Images
SN - 978-989-758-475-6
AU - Timme F.
AU - Kerdels J.
AU - Peters G.
PY - 2020
SP - 396
EP - 403
DO - 10.5220/0010107203960403
PB - SciTePress