Ehsan Nowroozi

Ehsan Nowroozi
Sabanci University · Faculty of Engineering and Natural Sciences

Post Doctoral Fellow
Senior IEEE Member, Postdoctoral Fellow, Sabanci University, Istanbul, Turkey https://enowroozi.com/

About

37
Publications
49,303
Reads
How we measure 'reads'
A 'read' is counted each time someone views a publication summary (such as the title, abstract, and list of authors), clicks on a figure, or views or downloads the full-text. Learn more
207
Citations
Introduction
Research interests include Adversarial Multimedia Forensics, Multimedia Forensics (in particular, image forensics), Machine/Deep Learning approaches applied to forensic problems. Website: https://www.math.unipd.it/~nowroozi/ Google Scholar: https://scholar.google.com/citations?user=C0bNkP8AAAAJ&hl=en
Additional affiliations
March 2021 - March 2022
University of Padova
Position
  • PostDoctoral
April 2020 - March 2021
Università degli Studi di Siena
Position
  • PostDoc Position
Description
  • Postdoctoral Fellow
October 2016 - October 2019
Università degli Studi di Siena
Position
  • PhD
Description
  • PhD
Education
October 2016 - October 2019
Università degli Studi di Siena
Field of study
  • PhD on Information Engineering and Science

Publications

Publications (37)
Research
Full-text available
Nowadays with advancement of technology, tampering of digital images using computer and advanced software packages like Photoshop has become a simple task. Many algorithms have been proposed to detect tampered images that have been kept developing. In this regard, verification of accuracy of image content and detection of manipulations in image reg...
Book
Full-text available
The rapid growth of image processing software programs and the development of digital cameras have led to a large number of distorted images with no clear trace. This has resulted in numerous demands for identifying automatic algorithms used for image forgery aimed at determining the authenticity of these images. Unfortunately, image editing tools...
Preprint
Full-text available
Malicious advertisement URLs pose a security risk since they are the source of cyber-attacks, and the need to address this issue is growing in both industry and academia. Generally, the attacker delivers an attack vector to the user by means of an email, an advertisement link or any other means of communication and directs them to a malicious websi...
Preprint
Full-text available
Recently, the popularity and wide use of the last-generation video conferencing technologies created an exponential growth in its market size. Such technology allows participants in different geographic regions to have a virtual face-to-face meeting. Additionally, it enables users to employ a virtual background to conceal their own environment due...
Preprint
Full-text available
In the past decades, the excessive use of the last-generation GAN (Generative Adversarial Networks) models in computer vision has enabled the creation of artificial face images that are visually indistinguishable from genuine ones. These images are particularly used in adversarial settings to create fake social media accounts and other fake online...
Article
Full-text available
Convolutional Neural Networks (CNNs) models are one of the most frequently used deep learning networks, and extensively used in both academia and industry. Recent studies demonstrated that adversarial attacks against such models can maintain their effectiveness even when used on models other than the one targeted by the attacker. This major propert...
Preprint
Full-text available
Deep Convolutional Neural Networks (CNN) models are one of the most popular networks in deep learning. With their large fields of application in different areas, they are extensively used in both academia and industry. CNN-based models include several exciting implementations such as early breast cancer detection or detecting developmental delays i...
Article
Full-text available
The possibility of carrying out a meaningful forensic analysis on printed and scanned im- ages plays a major role in many applications. First of all, printed documents are often associated with criminal activities, such as terrorist plans, child pornography, and even fake packages. Additionally, printing and scanning can be used to hide the traces...
Preprint
Full-text available
The possibility of carrying out a meaningful forensics analysis on printed and scanned images plays a major role in many applications. First of all, printed documents are often associated with criminal activities, such as terrorist plans, child pornography pictures, and even fake packages. Additionally, printing and scanning can be used to hide the...
Presentation
Full-text available
Last-generation GAN models allow to generate synthetic images which are visually indistinguishable from natural ones, raising the need to develop tools to distinguish fake and natural images thus contributing to preserve the trustworthiness of digital images. While modern GAN models can generate very high-quality images with no visible spatial arti...
Article
Full-text available
Image forensic plays a crucial role in both criminal investigations (e.g., dissemination of fake images to spread racial hate or false narratives about specific ethnicity groups) and civil litigation (e.g., defamation).Increasingly, machine learning approaches are also utilized in image forensics. However, there are also a number of limitations and...
Preprint
Full-text available
Image forensic plays a crucial role in both criminal investigations (e.g., dissemination of fake images to spread racial hate or false narratives about specific ethnicity groups) and civil litigation (e.g., defamation). Increasingly, machine learning approaches are also utilized in image forensics. However, there are also a number of limitations an...
Conference Paper
Full-text available
Last-generation GAN models allow to generate synthetic images which are visually indistinguishable from natural ones, raising the need to develop tools to distinguish fake and natural images thus contributing to preserve the trustworthiness of digital images. While modern GAN models can generate very high-quality images with no visibile spatial art...
Preprint
Full-text available
Last-generation GAN models allow to generate synthetic images which are visually indistinguishable from natural ones, raising the need to develop tools to distinguish fake and natural images thus contributing to preserve the trustworthiness of digital images. While modern GAN models can generate very high-quality images with no visible spatial arti...
Thesis
Full-text available
The use of machine-learning for multimedia forensics is gaining more and more consensus, especially due to the amazing possibilities offered by modern machine learning techniques. By exploiting deep learning tools, new approaches have been proposed whose performance remarkably exceed those achieved by state-of-the-art methods based on standard mach...
Article
Full-text available
Protecting image manipulation detectors against perfect knowledge attacks requires the adoption of detector architectures which are intrinsically difficult to attack. In this paper, we do so, by exploiting a recently proposed multiple-classifier architecture combining the improved security of 1-Class (1C) classification and the good performance ens...
Preprint
Full-text available
We investigate if the random feature selection approach proposed in [1] to improve the robustness of forensic detectors to targeted attacks, can be extended to detectors based on deep learning features. In particular, we study the transferability of adversarial examples targeting an original CNN image manipulation detector to other detectors (a ful...
Preprint
Recent studies have shown that Convolutional Neural Networks (CNN) are relatively easy to attack through the generation of so-called adversarial examples. Such vulnerability also affects CNN-based image forensic tools. Research in deep learning has shown that adversarial examples exhibit a certain degree of transferability, i.e., they maintain part...
Preprint
Recent studies have shown that Convolutional Neural Networks (CNN) are relatively easy to attack through the generation of so called adversarial examples. Such vulnerability also affects CNN-based image forensic tools. Research in deep learning has shown that adversarial examples exhibit a certain degree of transferability, i.e., they maintain part...
Poster
Full-text available
Detection of contrast adjustments in the presence of JPEG post processing is known to be a challenging task. JPEG post processing is often applied innocently, as JPEG is the most common image format, or it may correspond to a laundering attack, when it is purposely applied to erase the traces of manipulation. In this paper, we propose a CNN-based d...
Conference Paper
Full-text available
Detection of contrast adjustments in the presence of JPEG post processing is known to be a challenging task. JPEG post processing is often applied innocently, as JPEG is the most common image format, or it may correspond to a laundering attack, when it is purposely applied to erase the traces of manipulation. In this paper, we propose a CNN-based d...
Poster
Full-text available
Contrast Enhancement (CE) detection in the presence of laundering attacks, i.e. common processing operators applied with the goal to erase the traces the CE detector looks for, is a challenging task. JPEG compression is one of the most harmful laundering attacks, which has been proven to deceive most CE detectors proposed so far. In this paper, we...
Conference Paper
Full-text available
Contrast Enhancement (CE) detection in the presence of laundering attacks, i.e. common processing operators applied with the goal to erase the traces the CE detector looks for, is a challenging task. JPEG compression is one of the most harmful laundering attacks, which has been proven to deceive most CE detectors proposed so far. In this paper, we...
Preprint
Full-text available
Detection of contrast adjustments in the presence of JPEG postprocessing is known to be a challenging task. JPEG post processing is often applied innocently, as JPEG is the most common image format, or it may correspond to a laundering attack, when it is purposely applied to erase the traces of manipulation. In this paper, we propose a CNN-based de...
Research
Full-text available
Certificate of Attendance (EUSIPCO 2017), Kos Island, Greece
Conference Paper
Full-text available
In this paper, we present an adversary-aware double JPEG detector which is capable of detecting the presence of two JPEG compression steps even in the presence of heterogeneous processing and counter-forensic (C-F) attacks. The detector is based on an SVM classifier fed with a large number of features and trained to recognize the traces left by dou...
Article
Full-text available
Nowadays a variety of powerful software has been developed to manipulate digital images, that such software has caused the accuracy of digital medical documents undergo questioning. Yet, fortunately some algorithms have developed to diagnose fake medical images that undergo progress. Any change in digital images will raise unconventional statistica...
Book
Full-text available
http://ketab.org.ir/bookview.aspx?bookid=1964231
Thesis
Full-text available
Nowadays with technology progress, editing and tampering of digital images using computer and advanced software packages like Photoshop is a simple task. Many algorithms have been proposed to detect tampered images and are being improved. Generally, these algorithms fall into two categories: Active and Passive. Active methodologies, need to applian...
Conference Paper
Full-text available
In this paper, some new indexing methodologies and applications in Information Retrieval (IR) has been presented. Some new algorithms with high coverage of IR applications has be introduced by this paper. Main strategy is introducing and evaluating Information Retrieval basic applications and modulation. Some future directions in IR methodologies a...

Questions

Questions (53)
Question
If we have two backgrounds stacked on top of each other, how we can detect or reconstruct the first background In image or in video?
i appreciate it tell me if there is any paper, GitHub....
Question
I need to find TTL from Python and I found several codes but I don't know why TTL output in python different than TTL when I used ping www.google.com?
for example in python TTL = 45 or 226 ...but in command line terminal ping www.google.com TTL =118.
Do you know any python code that I obtain TTL which is matched with TTL in the command line?
Question
Dear Researchers,
`Since I am new in the field of internet security, I need your suggestion regarding the meaning of the following features.
We have DNS google.com or youtube.com, and so on, and I want to extract different features based on Lexical and Web Scrapped.
Lexical Features:
what is the meaning of the following features? Please write with an example.
1) different ratios (different ratios (number to length, alphabet to length) ?
2) hash?
3) distance between a number to an alphabet? (You can find the meaning of these features in the paper Feature Extraction Approach to Unearth Domain Generating Algorithms (DGAs) - Page 401)
4) English domain name, not English yet pronounceable domain names, uni-gram?
Web Scrapping:
we extract information of the queried domain name from the web using Python (You can find the meaning of these features in the paper Feature Extraction Approach to Unearth Domain Generating Algorithms (DGAs) - Page 403)).
1) Levenshtein distance (sq1,se2), what is seq2?
2) Typosquat process?
Thanks
Question
Dear Researchers,
`Since I am new in the field of internet security, I need your suggestion regarding the meaning of the following features.
We have DNS google.com or youtube.com, and so on, and I want to extract different features based on Lexical and Web Scrapped.
Lexical Features:
what is the meaning of the following features? Please write with an example.
1) different ratios (different ratios (number to length, alphabet to length) ?
2) hash?
3) distance between a number to an alphabet? (You can find the meaning of these features in the paper Feature Extraction Approach to Unearth Domain Generating Algorithms (DGAs) - Page 401)
4) English domain name, not English yet pronounceable domain names, uni-gram?
Web Scrapping:
we extract information of the queried domain name from the web using Python (You can find the meaning of these features in the paper Feature Extraction Approach to Unearth Domain Generating Algorithms (DGAs) - Page 403)).
1) Levenshtein distance (sq1,se2), what is seq2?
2) Typosquat process?
Thanks
Question
Dear All,
Id you are interested in the area my Adversarial Multimedia Forensics, myy PhD thesis now available on EURASIP database at the https://theses.eurasip.org/theses/859/machine-learning-techniques-for-image-forensics/
Thanks
Question
I read one paper and in the part of Model selection for the detection they considered Random Forest and for prediction they considered Decision tree.
I am wondering to knows each classifiers works for specific tasks, I mean prediction and detection? and which classifiers we can consider for detection or prediction?
Question
I read one paper and in the part of Model selection for the detection they considered Random Forest and for prediction they considered Decision tree.
I am wondering to knows each classifiers works for specific tasks, I mean prediction and detection? and which classifiers we can consider for detection or prediction?
Question
Since the input matrix is too big for training and I need to wait more than an hour I want to know how can I run through GPU? Also, my Cuda version is v-10.0.
I also try to use the Cupy-v100 ( https://github.com/murtazajafferji/svm-gpu ) library but it seems it does not work properly.
1) clf = GridSearchCV(SVC(class_weight='balanced',probability=True), tuned_parameters, cv=5, scoring=score)
clf.fit(train_data, train_label)
2)clf_full=SVC(kernel=clf.best_params_['kernel'],gamma=clf.best_params_['gamma'],C=clf.best_params_['C'],class_weight='balanced',probability=True) clf_full.fit(train_data, train_label)
Question
Since the input matrix is too big for training and I need to wait more than an hour I want to know how can I run through GPU? Also, my Cuda version is v-10.0.
I also try to use the Cupy-v100 ( https://github.com/murtazajafferji/svm-gpu ) library but it seems it does not work properly.
1) clf = GridSearchCV(SVC(class_weight='balanced',probability=True), tuned_parameters, cv=5, scoring=score)
clf.fit(train_data, train_label)
2)clf_full=SVC(kernel=clf.best_params_['kernel'],gamma=clf.best_params_['gamma'],C=clf.best_params_['C'],class_weight='balanced',probability=True) clf_full.fit(train_data, train_label)
Question
Dear Researchers,
It is my pleasure to know your opinions on my friend Yousef Safiran Thesis topic. With your help, he is going to Investigate the Effect of Marketing Strategies such as Product Marketing Strategies as well as Web-Vendor Marketing Strategies on Online Shopping Valuation in AMAZON with the Intermediate role of Risks, Customer Trust, Loyalty and Intention to purchase.
Please fill the below questionnaire,
Thank you so much in advance.
The University of Siena, Italy
Question
Dear All,
I need python for the co-occurrence matrix method between the color bands [R,G] , [R,B] , [G,B]....
Thanks
Question
Dear All,
I need python for the co-occurrence matrix method between the color bands [R,G] , [R,B] , [G,B]....
Thanks
Question
Why IFGSM attack always works but in deep neural network sometime, it happens. Something during simulations I found that gradient vanishing happened in deep neural networks in all of the case of epsilons, but if we decrease epsilons it seems work and problem of gradient vanishing approximately will be solved.
I need to complete intuitive regarding this topic also paper can help me for understanding about this question.
Thanks
Question
Dear All,
Based on I-FGSM (Iterative Fast Gradient Sign Attack) in FoolBox Library Attack, (https://foolbox.readthedocs.io/en/latest/) the attack try to cross the region (example "positive to negative" or "negative to positive") but when the gradient cross from the region the attack will be stopped but the attack near the border not go more inside the opposite region. I want to modify this part which attack goes more inside the region but I don't know which part of the code should be modified or function or class?
Ehsan
Question
Dear Researchers,
I want to know the parameter Gamma in one-class classifier (SVM) shows what things exactly?
Is it correct if parameter Gamma is large, then fewer support vectors are selected and finally the region becomes more spherical ???
Thanks
Question
Dear researchers,
In machine learning it is possible to use one-class-classification but I don't know it's possible also to do similar way using deep networks such as CNN or not?
If there is any references in this area please consider.
Thanks
Question
Dear Researchers,
I need several references (articles) regarding how to improve the security of deep networks specially CNN?
Thanks
Question
Dear Researchers,
I am more curious to know which differences between the Matlab JPEG Compression and also Photoshop JPEG Compression which means that I know standard quantization tables in both software compression are different but is there any difference between both them such as (e.g. block 8*8, DCT and Huffman coding and so on).
Please mention if there is any reference concerning this question.
Regards,
Ehsan

Network

Cited By

Projects

Projects (2)
Project
The goal of the present research is to: i) investigate the security of machine-learning tools for multimedia forensics, ii) develop counter-forensic attacks against state-of-the-art techniques, iii) develop a new class of security-enhanced multimedia forensic techniques based on machine learning.