Attack Type Agnostic Perceptual Enhancement Of Adversarial Images

Date

2020-07-19

Author

Aksoy, Bilgin
Temizel, Alptekin

Metadata

Show full item record

Item Usage Stats

213
views

0
downloads

Adversarial images are samples that are intentionally modified to deceive machine learning systems. They are widely used in applications such as CAPTHAs to help distinguish legitimate human users from bots. However, the noise introduced during the adversarial image generation process degrades the perceptual quality and introduces artificial colours; making it also difficult for humans to classify images and recognise objects. In this letter, we propose a method to enhance the perceptual quality of these adversarial images. The proposed method is attack type agnostic and could be used in association with the existing attacks in the literature. Our experiments show that the generated adversarial images have lower Euclidean distance values while maintaining the same adversarial attack performance. Distances are reduced by 5.88% to 41.27% with an average reduction of 22% over the different attack and network types.

URI

https://hdl.handle.net/11511/86059

Conference Name

International Workshop on Adversarial Machine Learning And Security (AMLAS), IEEE World Congress on Computational Intelligence (IEEE WCCI),19 Temmuz 2020

Collections

Graduate School of Informatics, Conference / Seminar

Suggestions

OpenMETU
Core

Well test model identification by artificial neural networks Kök, Mustafa Verşan (Informa UK Limited, 2000-01-01) The aim of this research is to investigate the performance of artificial neural networks computing technology, to identify preliminary well test interpretation model based on derivative plot. The approach is based on training the neural network simulator uses back-propagation as the learning algorithm for a predefined range of analytically generated well test response. The trained network is then requested to identify the well test identification model for test data, which is not used during training sessio...
Classification via ensembles of basic thresholding classifiers TOKSÖZ, Mehmet Altan; Ulusoy, İlkay (Institution of Engineering and Technology (IET), 2016-08-01) The authors present a sparsity-based algorithm, basic thresholding classifier (BTC), for classification applications which is capable of identifying test samples extremely rapidly and performing high classification accuracy. They introduce a sufficient identification condition (SIC) under which BTC can identify any test sample in the range space of a given dictionary. By using SIC, they develop a procedure which provides a guidance for the selection of threshold parameter. By exploiting rapid classification...
A new neural network approach to the target tracking problem with smart structure Caylar, Selcuk; Leblebicioğlu, Mehmet Kemal; Dural, Gülbin (2006-12-01) The algorithm presented in this paper, namely the modified neural multiple source tracking algorithm (MN-MUST) is the modified form of the recently published work, a NN algorithm, the neural multiple-source tracking (N-MUST) algorithm, was presented for locating and tracking angles of arrival from multiple sources. MN-MUST algorithm consists of three stages that are classified as the detection, filtering and DoA estimation stages. In the first stage a number of radial basis function neural networks (RBFNN) ...
Estimation of Noise Model Parameters for Images Taken by a Full-frame Hyperspectral Camera DEMİRKESEN, Can; Leloğlu, Uğur Murat (2015-09-23) Noise has to be taken into account in the algorithms of classification, target detection and anomaly detection. Recent studies indicate that noise estimation is also crucial in subspace identification of Hyper Spectral Images (his). Several techniques were proposed for noise estimation including: multiple linear regression based techniques, spectral unmixing and remixing etc. The noise in HSI is widely accepted to be a spatially stationary random process. But the variance of the noise varies from one wavele...
Signature Based Vegetation Detection on Hyperspectral Images Özdemir, Okan Bilge; Soydan, Hilal; Çetin, Yasemin; Düzgün, Hafize Şebnem (2015-05-19) In this study, the contribution of utilizing hyperspectral unmixing algorithms on signature based target detection algorithms is studied. Spectral Angle Mapper (SAM), Spectral Matched Filter (SMF) and Adaptive Cosine Estimator (ACE) algorithms are selected as target detection methods and the performance change related to the target spectral acquisition is evaluated. The spectral signature of the desired target, corn, is acquired from ASD hyperspectral library as well as from the hypespectral unmixing endmem...

Citation Formats

B. Aksoy and A. Temizel, “Attack Type Agnostic Perceptual Enhancement Of Adversarial Images,” presented at the International Workshop on Adversarial Machine Learning And Security (AMLAS), IEEE World Congress on Computational Intelligence (IEEE WCCI),19 Temmuz 2020, Glasgow, İskoçya, 2020, Accessed: 00, 2021. [Online]. Available: https://hdl.handle.net/11511/86059.