Comparative analysis of feature extraction and fusion for blind authentication of digital images using chroma channels
Introduction
Nowadays, we live in an era where different technologies are used to automatically share, access and process different types of information. Visual online content is among the most significant information that poses several challenges. The Chinese proverb described the power of the picture with “A picture is worth a thousand of words”. However, due to the availability of powerful editing tools, images can easily be manipulated in a way that is hard to realize with bare eyes. Therefore, seeing is no longer believing. Image tampering was known over years,1 but it became a pandemic problem with the widespread of technologies. Fig. 1 shows examples of two common forgery techniques: splicing (combining parts of different images) and copy–move (duplicating regions of the same image). Fig. 1(a) shows a fake spliced image, which has never happened, where Senator John Kerry is speaking to the crowd with a Hollywood actress, Jane Fonda, whom was considered a traitor by many Americans during the Vietnam war. It was manipulated and launched during the presidential election campaign to raise doubts on John’s patriotism. The other example shown in Fig. 1(b) is a forged image for launching missiles by the Iranian government during the Iraq war by replicating other parts.2
Image manipulation aims at altering an image content without leaving traceable footprints using various methods to achieve some desired results for legitimate or illegal reasons. For instance, manipulation techniques are not new, and are acceptable in some cases such as photomontage and visual effects in movies, medical imaging, and journalism. However, tampered images with intimidating effects have been immorally flooding the Internet to provide false information for political, financial or social gains. Recently, several instances have been witnessed with the emergence of the Deepfake technology to synthesize apparently realistic visual content [1].
Image forensic is an area where images are analyzed to investigate changes or tamper. Different techniques have been proposed for detecting tampered images, which are categorized as active or passive/blind [2]. Active approaches need some prior information such as embedded digital signatures or watermarks to be used for image-authenticity check. This limits their application because embedded information may affect the image quality and/or require specially-equipped cameras [3]. On the other hands, passive approaches, which are the scope of this study, do not need information other than the manipulated image itself. Passive detection of image forgery is a crucial research area as the volume of online media is dramatically rising making distinction between original and forged images a daunting task.
There are several efforts in the literature to passively detect image forgery [4], [5]. From a pattern recognition perspective, this problem is a binary classification problem aiming to determine the right category for a given image, i.e. is it original or forged? A major component in this process is extraction of relevant features to represent each image by a feature vector in a low-dimensional space. Several methods have been proposed based on texture representation and image transformation. Unfortunately, there is no comprehensive study to quantitatively compare and benchmark these methods. This challenge is addressed in this study by presenting a framework for a comprehensive analysis of various texture-based feature extraction methods for image forgery detection. Since each method has its own merits and demerits to extract local and global features, the proposed framework enables multimodal feature extraction by combining a number of methods in a variety of pipelines to extract and merge higher level abstractions from diverse sources. Two benchmark datasets (CASIA v1.0 and CASIA v2.0) [6] are used to evaluate and statistically compare the various approaches considered in this study.
The rest of the paper is organized as follow: Section 1 provides an introduction, defines the research problem and lists the research objectives. Section 2 briefly gives background and reviews related work. Section 3 describes the details of the research methodology. Section 4 describes the conducted experiments and discusses the results of various approaches on two publicly-available datasets. The last section concludes the work and highlights some research directions for future work.
Section snippets
Background and related work
Different techniques have been proposed in the literature for passive image forgery detection. There are a number of surveys published on various aspects of this field, e.g. [2], [3], [7], [8], [9]. Fig. 2 shows a taxonomy of various categories of image forgery detection techniques. Passive detection approaches are categorized into type-dependent and type-independent methods. While type-dependent methods can be used to detect either copy–move or splicing, type-independent methods can be used to
Methodology
We developed a comprehensive framework for image forgery detection. The typical layout shown in Fig. 4 has four main stages, which will be described with more details in the following subsections. Feature extraction is one of the fundamental steps to differentiate between authentic and forged images. These features convert the visual information into statistical numbers to reduce the large amount of data. Initially these features can be large in size and redundant which needs a large amount of
Experiments and results
In this section, the results of several experiments are reported to compare various approaches using two benchmark image datasets (CASIA v1.0 and CASIA v2.0) containing both spliced and copy–move tampered images. CASIA v1.0 dataset consists of 800 authentic and 921 tampered images in two different sizes 384 × 256 and 256 × 384 in JPEG format. The images include various categories of natural and textural scenes like plants, animals and architectures. On contrast, CASIA v2.0 is an extended
Conclusions
Image forensic is a crucial research area in the current technological era as more fake images are shared via electronic media. In this work, we handle the problem of detecting image forgery using various texture-based approaches. We presented a comprehensive framework that combines multiple methods and enables various feature extractors to be compared. We focused on LBP, MSLBP, CSLBP, SLBP, WLD, LPQ, MSLPQ and DCT and their combinations. SVM with Polynomial and RBF kernels were utilized for
CRediT authorship contribution statement
Atif Shah: Methodology, Software, Investigation, Validation, Writing - original draft, Visualization. El-Sayed M. El-Alfy: Conceptualization, Methodology, Writing - review & editing, Validation, Visualization, Supervision, Project administration.
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.
Acknowledgment
The authors would like to thank King Fahd University of Petroleum and Minerals, Saudi Arabia for support during this work.
References (80)
- et al.
Digital image forgery detection using passive techniques: A survey
Digit. Investig.
(2013) - et al.
A bibliography on blind methods for identifying image forgery
Signal Process., Image Commun.
(2010) - et al.
A bibliography of pixel-based blind image forgery detection techniques
Signal Process., Image Commun.
(2015) - et al.
Copy-move forgery detection: survey, challenges and future directions
J. Netw. Comput. Appl.
(2016) - et al.
A survey on image tampering and its detection in real-world photos
J. Vis. Commun. Image Represent.
(2019) - et al.
A generic passive image forgery detection scheme using local binary pattern with rich models
Comput. Electr. Eng.
(2017) - et al.
Quaternion pseudo-zernike moments combining both of rgb information and depth information for color image splicing detection
J. Vis. Commun. Image Represent.
(2017) - et al.
Digital image modification detection using color information and its histograms
Forensic Sci. Int.
(2016) - et al.
An improved parameter estimation scheme for image modification detection based on dct coefficient analysis
Forensic Sci. Int.
(2016) - et al.
Digital image splicing detection based on markov features in dct and dwt domain
Pattern Recognit.
(2012)