A critical literature survey and prospects on tampering and anomaly detection in image data
Introduction
In the last decade, digital images have thoroughly replaced conventional photographs. With that advancement, the usage of digital images in a computational forensics field has become more apparent. The significant development in digital image processing, as much as it has helped in the evolution of many new techniques in forensic studies, has also made image tampering easy. In this sense, Image security has become a critical issue in all field that makes use of digital images. Tampered images have long been contents of forensic studies; for example, images of criminals, images of crime scenes, biometric images, etc [1], [2], [3].
A digital image may be defined as a numerical representation of any scene. Thus, manipulating such images has become an easy task even for non-specialists, due to simplified tools available on any device, e.g., as smartphones and tablets. In this context, elements are combined to create a unique image with the potential to convince even the most experienced set of eyes [4]. The manipulation of images through forgery can impact the perception a viewer has of the depicted scene. Therefore, identifying unknown sources’ image authenticity has become of paramount importance, in the absence of any prior digital watermarking or other authentications techniques [5], [6].
Although the detection of image manipulation may present some similarities with semantic object detection, they differ in the context of the objective, since image manipulation detection focuses more on tampering artifacts instead of image content, which demands more sophisticated approaches. Therefore, distinguishing tempered images poses an increasing challenge, which requires the development of new techniques, as well as the enhancement of the ones developed so far, for proper identification of the image tampering operation performed over the models [7], [8].
Therefore, this work provides an overview of different research findings, and propose solutions to improve image anomaly detection (IAD) in the context of images tampering, using machine learning and other intelligent techniques, such as evolutionary computation, neural networks, fuzzy logic, and Bayesian reasoning. Moreover, it also presents some highlights concerning image tampering, as well as the approaches proposed to deal with the issues mentioned above through machine learning methods.
The organization of this work is presented as follows. Section 2 introduces a collection of researches related to traditional machine learning-based algorithms employed for image tampering detection (ITD) and further discusses the relevant contributions associated with anomaly detection (AD) methods applied to the same context. Section 3 presents the most commonly employed datasets for the task, as well as the intelligent techniques adopted in the proposed experiments, while Section 4 discusses the opportunities and challenges faced by the works considered in this survey. Finally, Section 5 states conclusions and a discussion about the future possibilities for research in image tampering.
In the last years, the interest of researchers towards image tampering detection and image anomaly detection increased considerably, due to their impact potential over different human actions. In this context, many works focused on developing feasible solutions to detected tampering in several image-related approaches, e.g., handwriting forgery [9], face spoofing [10], and machine learning-based image classifiers attack [11]. Moreover, image manipulation tools have been coupled to mobile devices, thus becoming popular on people daily’s life due to their simplicity, and also contributing to the emergence of image tampering.
Therefore, improvements towards image security become crucial, specifically considering image tampering detection [12], [13]. The field provides a brand new ground for research, whose demand grows to the same extent as manipulating tools become easier to handle and more popular, implying in the increasing number of fake images. Furthermore, in general, these fake images carry a high level of confidence, implying severe consequences to people and, at a higher level, even to governments and institutions [14].
However, as far as we are concerned, no work has presented an in-depth view regarding machine learning and intelligent system applications in the context of image tampering detection, which ends up being the main contribution of this survey.
This survey aims at providing an overview of the research progress related to security-related issues in image tampering detection. The scope of this work discusses methods of image tampering detection based on machine learning and evolutionary computation, among other intelligent systems.
The idea of this review is to provide information available on the current literature, as well as to be a new source for researchers interested in image detection and security issues. Additionally, it presents a clear vision of the possible challenges of current research and the new research guidelines.
Section snippets
Related works
The works presented in this survey highlights specific methods for image anomaly detection, focusing on a variety of autonomous intelligent systems. Besides, it also presents a sort of algorithms’ variations, which were improved for better performance in the context. Fig. 1 depicts a general pipeline applied for image anomaly detection employing intelligent techniques. Notice this architecture considers datasets collected and properly prepared for the proposed tasks, i.e., the data is processed
Methods and datasets
This section discusses the methodologies and datasets used in recent work focusing on Anomaly Detection for Hyperspectral Imagery.
In this context, the PCA skeleton kernel algorithm proposed by Olson et al. [85] performed very well over several methods and datasets. The method has proved to be reasonably robust to the variability of subsamples selection. Further, experiments were conducted over “The Hyperspectral Digital Imaging Collection Experiment (HYDICE)”, a dataset composed of
Discussion and open issues
The number of easy-to-use image editors has swiftly increased in the last decade, contributing to image manipulation even by lay users. However, such a phenomenon opens space for several risks concerning image security, such as real images tampering. Therefore, robust approaches and enhanced methods are desirable to avoid such incidents and provide more stable detection. Further, more studies are required to reduce the risk of potential threats in confidentiality and privacy, for instance,
Conclusions
This work highlighted the most significant works developed in the last years in the context of seam carved, tampered, and anomalous imagery detection through machine learning techniques. Besides, it also presents some advances regarding new tampering techniques, which poses new challenges to the field and demands a parallel advance in the detection techniques.
The works surveyed in this paper discusses, mainly, the concerns and efforts made by the scientific community and the industry towards
CRediT authorship contribution statement
Kelton A.P. da Costa: Conceptualization, Methodology, Software, Writing - original draft. João P. Papa: Conceptualization, Writing - review & editing. Leandro A. Passos: Conceptualization, Writing - review & editing. Danilo Colombo: Conceptualization, Writing - review & editing. Javier Del Ser: Validation, Writing - review & editing. Khan Muhammad: Formal analysis, Investigation, Validation, Writing - review & editing, Supervision. Victor Hugo C. de Albuquerque: Formal analysis, Investigation,
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
The authors are grateful to Fundação de Amparo á Pesquisa do Estado de São Paulo (FAPESP), Brazil grants #2017/22905-6, #2013/07375-0, #2014/12236-1, and #2016/19403-6 and by the Brazilian National Council for Research and Development (CNPq) via grants -8, -6, -1,-7 and -6, Petrobras, Brazil (grant-0) for their financial support and Basque Government, Spain for its funding support through the ELKARTEK and EMAITEK funding
References (130)
- et al.
A survey on image tampering and its detection in real-world photos
J. Vis. Commun. Image Represent.
(2019) - et al.
Image splicing detection based on Markov features in QDCT domain
Neurocomputing
(2017) - et al.
Tamper detection of social media images using quality artifacts and texture features
Forensic Sci. Int.
(2019) - et al.
A probabilistic Hough transform
Pattern Recogn.
(1991) - et al.
Nonoverlapping blocks based copy-move forgery detection
Secur. Commun. Netw.
(2018) - et al.
Novel image authentication with tamper localization and self-recovery in encrypted domain based on compressive sensing
Secur. Commun. Netw.
(2018) - et al.
Perceptual hashing-based image copy-move forgery detection
Secur. Commun. Netw.
(2018) - et al.
Internet of Things: A survey on machine learning-based intrusion detection approaches
Comput. Netw.
(2019) - et al.
Context-dependent image quality assessment of JPEG compressed mars science laboratory mastcam images using convolutional neural networks
Comput. Geosci.
(2018) - et al.
Unexpected item in the bagging area: Anomaly detection in X-ray security images
IEEE Trans. Inf. Forensics Secur.
(2018)