Comparative analysis of feature extraction and fusion for blind authentication of digital images using chroma channels

https://doi.org/10.1016/j.image.2021.116271Get rights and content

Highlights

  • Evaluate and benchmark different models of feature extraction to detect digitally-altered images.

  • Combine various features utilizing single and multi-scale representations to enhance recognition accuracy.

  • Intensively compare and statistically analyze different models integrating the three color channels.

  • Assess the impact on the performance of two feature reduction methods using PCA and LLP.

Abstract

Blind authentication is one of the challenging techniques which has attracted considerable attention with the increasing security hacks on digital images. This paper evaluates different models based on feature extraction to detect digitally-altered images. Various feature-extraction methods have been investigated and compared including LBP, MSLBP, CSLBP, SLBP, WLD, LPQ, MSLPQ and DCT. Moreover, we propose a number of ways to combine a variety of features utilizing single and multi-scale representations of images. For classification, SVM is employed with different kernels to identify forged and authentic images. To evaluate the effectiveness of the investigated models, several experiments have been conducted using k-fold cross-validation and computed performance measures for two benchmark image tampering datasets (CASIA v1.0 and CASIA v2.0). Additionally, we have conducted statistical analysis for the top-ten models and the results confirmed that the best models for CASIA v1.0 and CASIA v2.0 are MSLBP-DCT and MSLPQ, respectively. Further improvements have been achieved by integrating features from the three color channels (Y, Cb and Cr) with and without feature reduction using PCA and LLP. In this case, the results show that MSLPQ-DCT achieved a better accuracy of 98.56% on CASIA v1.0 with 1020 features and MSLPQ achieved a slightly better accuracy of 97.4% on CASIA v2.0 with 1536 features.

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)

  • MahmoodT. et al.

    Copy–move forgery detection through stationary wavelets and local binary pattern variance for forensic analysis in digital images

    Forensic Sci. Int.

    (2017)
  • HayatK. et al.

    Forgery detection in digital images via discrete wavelet and discrete cosine transforms

    Comput. Electr. Eng.

    (2017)
  • OjalaT. et al.

    A comparative study of texture measures with classification based on featured distributions

    Pattern Recognit.

    (1996)
  • HeikkiläM. et al.

    Description of interest regions with local binary patterns

    Pattern Recognit.

    (2009)
  • DavarzaniR. et al.

    Copy-move forgery detection using multiresolution local binary patterns

    Forensic Sci. Int.

    (2013)
  • KwokA.O. et al.

    Deepfake: a social construction of technology perspective

    Curr. Issues Tour.

    (2020)
  • FaridH.

    Image forgery detection

    IEEE Signal Process. Mag.

    (2009)
  • AnsariM.D. et al.

    Pixel-based image forgery detection: A review

    IETE J. Educ.

    (2014)
  • PhamN.T. et al.

    Hybrid image-retrieval method for image-splicing validation

    Symmetry

    (2019)
  • ZampoglouM. et al.

    Detecting image splicing in the wild (web)

  • D. Cozzolino, D. Gragnaniello, L. Verdoliva, Image forgery localization through the fusion of camera-based,...
  • ZhangY. et al.

    Blind forensics of median filtering based on markov statistics in median-filtered residual domain

  • FridrichJ. et al.

    Rich models for steganalysis of digital images

    IEEE Trans. Inf. Forensics Secur.

    (2012)
  • BahramiK. et al.

    Blurred image splicing localization by exposing blur type inconsistency

    IEEE Trans. Inf. Forensics Secur.

    (2015)
  • ZhaoX. et al.

    Detecting digital image splicing in chroma spaces

  • MayerO. et al.

    Accurate and efficient image forgery detection using lateral chromatic aberration

    Trans. Inf. Forensics Secur.

    (2018)
  • EmamM. et al.

    A robust detection algorithm for image copy-move forgery in smooth regions

  • ChristleinV. et al.

    An evaluation of popular copy-move forgery detection approaches

    IEEE Trans. Inf. Forensics Secur.

    (2012)
  • ChenB. et al.

    Fractional quaternion cosine transform and its application in color image copy-move forgery detection

    Multimedia Tools Appl.

    (2019)
  • GhoneimA. et al.

    Medical image forgery detection for smart healthcare

    IEEE Commun. Mag.

    (2018)
  • ChouC.-L. et al.

    Copy-move forgery detection based on local gabor wavelets patterns

  • Abdel-BassetM. et al.

    2-levels of clustering strategy to detect and locate copy-move forgery in digital images

    Multimedia Tools Appl.

    (2018)
  • PrakashC.S. et al.

    Image manipulation detection using harris corner and anms

  • ChenB. et al.

    Fast computation of sliding discrete tchebichef moments and its application in duplicated regions detection

    IEEE Trans. Signal Process.

    (2015)
  • FuD. et al.

    Detection of image splicing based on hilbert-huang transform and moments of characteristic functions with wavelet decomposition

  • X. Li, T. Jing, X. Li, Image splicing detection based on moment features and hilbert-huang transform, in: IEEE...
  • DongJ. et al.

    Run-length and edge statistics based approach for image splicing detection

  • W. Chen, Y.Q. Shi, W. Su, Image splicing detection using 2-d phase congruency and statistical moments of characteristic...
  • SchaeferG. et al.

    Ucid: an uncompressed color image database

  • AlkawazM.H. et al.

    Detection of copy-move image forgery based on discrete cosine transform

    Neural Comput. Appl.

    (2018)
  • View full text