Blind image splicing detection via noise level function

https://doi.org/10.1016/j.image.2018.07.012Get rights and content

Highlights

  • Investigating the influence of textures and edges on noise variance estimation.

  • Constructing noise level function to reflect the relationship between noise variance and sharpness.

  • A novel framework for image splicing detection based on noise level function is proposed.

Abstract

With the wide use of sophisticated image editing tools, digital image manipulation becomes very convenient and easy, which makes the detection of image tampering operations significant. Image splicing, which refers to a selected region from an image is cut and pasted onto another image, is a ubiquitous image tampering operation. In this paper, based on the finding that the textures and edges can affect the local estimation of noise variance, we propose a novel image splicing detection method via noise level function. Specifically, the test image is first segmented into non-overlapping blocks and the noise variance and the sharpness of each block are calculated. Then, the noise level function, which reflects the relationship between the noise variances and the sharpness of blocks is estimated. The goal of this function is to generate a distance map to describe the minimum distances from the blocks to the fitted curve. The blocks which not be constrained by this estimated noise level function will be distinct in the distance map so that they can be regarded as spliced regions. Finally, contextual information is additionally used to refine the detection result. Extensive experiments demonstrate the superiority of our proposed method in both quantitative and qualitative metrics when compared with the state-of-the-art approaches.

Introduction

With the rapid growth of image acquisition device and the popularity of social media, digital images have become our major information carrier, which leads to the fact that digital images are more and more widely utilized to support important decisions in our daily life. This is specially true in the applications related to law enforcement, military, commercial and scientific research. However, with the wide use of sophisticated image editing software, modifying the content of an image without leaving obvious visual traces has become easier than before. Therefore, it is very important to develop robust tampering detection tools to validate the authenticity of images. Although an embedded watermarking [1] or signature [2] can be used to verify the authenticity of an image, most digital images used in our daily life do not have either. As a consequence, blind digital image forensics technologies, which aim at verifying the originality and authenticity of digital images without any a prior knowledge, have absorbed considerable attention.

Over the past few years, many efforts have been devoted to develop various blind image forensic technologies [[3], [4], [5], [6]]. The existing techniques can be roughly classified into three categories according to different forensics features. The first category includes the techniques based on detecting the absence of special artifacts introduced during image acquisition process such as lateral chromatic aberration (LCA) [7], sensor pattern noise (SPN) [8], color filter array (CFA) interpolation [9], camera response function (CRF) [10], image quality degradation [11] and so on. The second category contains the techniques based on detecting the special traces left by the tampering operations such as contrast enhancement [12], double JPEG compression [[13], [14], [15]], blur inconsistency [16], seam carving [17], image scaling [18], image sharpening [19], noise level inconsistencies [20] and merged operation [21]. The last category covers the techniques based on semantic level such as lighting consistency [22] and shadow consistency [23].

Noise is inevitably introduced into an image during image acquisition pipeline and due to the inherent characteristics of each individual camera sensor, the noise variances are different between cameras. Ideally, the amount of the noise in an authentic digital image varies slightly across the entire image. Based on this assumption, by using a tool to blindly estimate the region-based noise variances in an image, spliced image regions can be identified through the inconsistencies of local noise variances [[20], [24], [25]]. However, this ideal assumption may not be true in practice. We find that the regions owning complex textures and sharp edges usually lead to an inaccurate estimation of noise variance, which decreases the performance of the following steps.

In this paper, we propose a novel blind image splicing detection method. Based on the finding that the textures and edges can affect the estimation of noise variance in an image, we construct a noise level function, which reflects the relationship between noise variance and sharpness of image blocks via polynomial curve fitting. The goal of this function is to generate a distance map, which can help people to make decision visually. The blocks which not be constrained by this estimated noise level function will be distinct in the distance map so that they can be regarded as spliced regions. A refinement scheme based on contextual information is utilized to provide the final detection result. Extensive experiments demonstrate the superiority of our proposed method in both quantitative and qualitative metrics when compared with the state-of-the-art approaches.

The remainder of this paper is organized as follows. The next section summarizes the existing approaches related to the topic of this paper. In Section 3, we describe our proposed image splicing detection method in detail. The experimental results are presented and discussed in Section 4. Finally, we draw the conclusions and the future work in Section 5.

Section snippets

Related work

Among digital image forensics technologies, the inconsistencies of image noise variances have been widely used for splicing detection. Generally, these methods share a common two-step pipeline. The first step is estimating noise variances across different regions in an image. In the second step, a block merging or clustering algorithm is utilized to separate the spliced region from the authentic part based on the inconsistencies of noise variances.

In [26], the noise variance of each image block

Proposed method

All the methods mentioned above are based on the assumption that the noise variances are similar across the authentic image and in case of image splicing, the noise variances of the spliced regions are typically different from the authentic part. However, this basic assumption may not be true as the existing local blind noise estimation methods are not always accurate in dealing with the regions owning complex textures and sharp edges. In practice, we find that the estimated noise variance of

Experiments

In this section, extensive experiments in both quantitative and qualitative metrics are given in comparison with some related noise variances inconsistencies based algorithms [[20], [24], [25]]. In the following, we note these three comparison methods as IVC, MMSEC and IJCV for short.

Conclusion

In this paper, we present a novel blind image splicing detection approach. Based on the finding that textures and edges can affect the local estimation of noise variance, we construct a noise level function to reflect the relationship between noise variance and sharpness of image blocks via polynomial curve fitting. The blocks which not be constrained by this estimated noise level function can be regarded as spliced regions. A refinement scheme with contextual information is utilized to give

References (37)

  • StammM.C. et al.

    Information forensics: An overview of the first decade

    IEEE Access

    (2013)
  • T. Gloe, K. Borowka, A. Winkler, Efficient estimation and large-scale evaluation of lateral chromatic aberration for...
  • KangX. et al.

    Enhancing source camera identification performance with a camera reference phase sensor pattern noise

    IEEE Trans. Inf. Forensics Secur.

    (2012)
  • FerraraP. et al.

    Image forgery localization via fine-grained analysis of CFA artifacts

    IEEE Trans. Inf. Forensics Secur.

    (2012)
  • HsuY.-F. et al.

    Camera response functions for image forensics: An automatic algorithm for splicing detection

    IEEE Trans. Inf. Forensics Secur.

    (2012)
  • P. Yang, R. Li, R. Ni, Y. Zhao, Recaptured image forensics based on quality aware and histogram feature, in:...
  • CaoG. et al.

    Contrast enhancement-based forensics in digital images

    IEEE Trans. Inf. Forensics Secur.

    (2014)
  • WangW. et al.

    Exploring DCT coefficient quantization effects for local tampering detection

    IEEE Trans. Inf. Forensics Secur.

    (2014)
  • Cited by (29)

    • Magnifying multimodal forgery clues for Deepfake detection

      2023, Signal Processing: Image Communication
    • Coarse-to-fine-grained method for image splicing region detection

      2022, Pattern Recognition
      Citation Excerpt :

      Pun et al. [15] reported an image splicing detection method based on multi-scale noise estimation, in which the noise level function was used to reflect the relationship between the noise level and brightness at different scales; the superpixel blocks that were not constrained by the noise level function were considered tampered regions. In [16], the authors used sharpness instead of the brightness used in the study in [15] and proposed an improved method. Noise-based image splicing detection methods typically provide good detection accuracy, but limitations remain.

    View all citing articles on Scopus
    View full text