Improving EfficientNet for JPEG Steganalysis

Yousfi, Yassine; Butora, Jan; Fridrich, Jessica; Tsang, Clément Fuji

doi:10.1145/3437880.3460397

Cited by 37 publications

(9 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…We relied on the U-Net model reported in [70], which allows to use various CNNs as backbones for the encoderdecoder structure. Our choice fell on the EfficientNetB0 model [71], a network of the EfficientNet family which proved extremely handy and recently found a discrete success both in multimedia forensics [72]- [74] and in overhead imagery analysis [75]. We considered an EfficientNetB0 as encoder, and another one as decoder.…”

Section: ) U-net Mask Estimatormentioning

confidence: 99%

Amplitude SAR Imagery Splicing Localization

et al. 2022

View full text Add to dashboard Cite

Synthetic Aperture Radar (SAR) images are a valuable asset for a wide variety of tasks. In the last few years, many websites have been offering them for free in the form of easy to manage products, favoring their widespread diffusion and research work in the SAR field. The drawback of these opportunities is that such images might be exposed to forgeries and manipulations by malicious users, raising new concerns about their integrity and trustworthiness. Up to now, the multimedia forensics literature has proposed various techniques to localize manipulations in natural photographs, but the same problem has never been investigated on SAR images. Forensics methods developed for natural photographs are not guaranteed to succeed on SAR images, as their generation pipeline is completely different from that of digital pictures. In this paper, we investigate the problem of localizing splicing attacks in amplitude SAR imagery. Our goal is to identify the pixels of an amplitude SAR image that have been copied and pasted from another image for malicious purposes, considering also that the attacker might have applied some editing to conceal this manipulation. To do so, we leverage a Convolutional Neural Network (CNN) to extract a fingerprint highlighting inconsistencies in the processing traces of the analyzed input. Then, we examine this fingerprint to produce a binary tampering mask indicating the pixel region under splicing attack. Results show that our proposed method, tailored to the nature of SAR signals, provides better performances than state-of-the-art forensic tools developed for natural images.

show abstract

Section: ) U-net Mask Estimatormentioning

confidence: 99%

Amplitude SAR Imagery Splicing Localization

et al. 2022

View full text Add to dashboard Cite

show abstract

“…The first set of classifiers F is available to Alice, who runs the min max protocol. Depending on the experiments, in this work, this set might contain either all classifiers with XuNet architecture [23] (differing in weights) noted F = {XuNet}, or all classifiers with XuNet, SrNet [24] and EfficientNet [25] architectures (denoted by…”

Section: Global Evaluationmentioning

confidence: 99%

“…It uses mini-batches containing 8 cover-stego pairs. The implementation of EfficientNet follows recommendations from [25]. We use version B0 and remove the stride in the first layer.…”

Section: Global Evaluationmentioning

confidence: 99%

Backpack: A Backpropagable Adversarial Embedding Scheme

Bernard

Bas

Klein

et al. 2022

IEEE Trans.Inform.Forensic Secur.

View full text Add to dashboard Cite

“…Importante destacar que os testes realizados quando a SRNET foi proposta utilizavam bases como BOSS, BOWS e BOWS-2. Nos experimentos mais recentes descritos pelos autores da SRNET [26][25] também tem sido utilizado o AdamW.…”

Section: Srnetunclassified

Efficientnets Aplicadas à Esteganálise Em Imagens Digitais

Albuquerque

Rodrigues

Ferrucio

et al. 2022

REPA

View full text Add to dashboard Cite

Diversas arquiteturas CNN com propósito específico para esteganálise foram desenvolvidas e atingiram o estado-da-arte superando os modelos anteriores que se baseavam nas etapas de extração de características e classificação. Novos conjuntos de dados de imagens foram propostos diferenciando-se dos anteriores pela quantidade de instâncias e a variação de características importantes como fator de qualidade e a carga útil (payload) de mensagem escondida em imagens. Além disso, novas arquiteturas de propósito geral têm se mostrado aplicáveis no âmbito da esteganálise e se beneficiam de transfer learning para acelerar o treinamento. Este trabalho aborda o treinamento da Steganalysis Residual Network (SRNET) com inicialização aleatória dos pesos e realiza a comparação de desempenho entre as arquiteturas de CNN Efficientnet e Efficientnetv2, com este último sendo 32% mais rápido que a EfficientnetB4, para cada época de treinamento. Por fim, também é apresentado um experimento envolvendo treinamentos sucessivos entre a imagem de cobertura e suas respectivas estego-imagens.

show abstract

Improving EfficientNet for JPEG Steganalysis

Cited by 37 publications

References 14 publications

Amplitude SAR Imagery Splicing Localization

Amplitude SAR Imagery Splicing Localization

Backpack: A Backpropagable Adversarial Embedding Scheme

Efficientnets Aplicadas à Esteganálise Em Imagens Digitais

Contact Info

Product

Resources

About