English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Deep learning offers state of the art solutions for image recognition.However, deep models are vulnerable to adversarial perturbations in images thatare subtle but significantly change the model's prediction. In a white-boxattack, these perturbations are generally learned for deep models that operateon RGB images and, hence, the perturbations are equally distributed in the RGBcolor space. In this paper, we show that the adversarial perturbations prevailin the Y-channel of the YCbCr space. Our finding is motivated from the factthat the human vision and deep models are more responsive to shape and texturerather than color. Based on our finding, we propose a defense againstadversarial images. Our defence, coined ResUpNet, removes perturbations onlyfrom the Y-channel by exploiting ResNet features in an upsampling frameworkwithout the need for a bottleneck. At the final stage, the untouchedCbCr-channels are combined with the refined Y-channel to restore the cleanimage. Note that ResUpNet is model agnostic as it does not modify the DNNstructure. ResUpNet is trained end-to-end in Pytorch and the results arecompared to existing defence techniques in the input transformation category.Our results show that our approach achieves the best balance between defenceagainst adversarial attacks such as FGSM, PGD and DDN and maintaining theoriginal accuracies of VGG-16, ResNet50 and DenseNet121 on clean images. Weperform another experiment to show that learning adversarial perturbations onlyfor the Y-channel results in higher fooling rates for the same perturbationmagnitude.

Adversarial Perturbations Prevail in the Y-Channel of the YCbCr Color Space