Pasquale Ferrara

Image Forgery Localization via Fine-Grained Analysis of CFA Artifacts

In this paper, a forensic tool able to discriminate between original and forged regions in an image captured by a digital camera is presented. We make the assumption that the image is acquired using a Color Filter Array, and that tampering removes the artifacts due to the demosaicking algorithm. The proposed method is based on a new feature measuring the presence of demosaicking artifacts at a local level, and on a new statistical model allowing to derive the tampering probability of each 2 × 2 image block without requiring to know a priori the position of the forged region. Experimental results on different cameras equipped with different demosaicking algorithms demonstrate both the validity of the theoretical model and the effectiveness of our scheme.

Reverse engineering of double compressed images in the presence of contrast enhancement

A comparison between two forensic techniques for the reverse engineering of a chain composed by a double JPEG compression interleaved by a linear contrast enhancement is presented here. The first approach is based on the well known peak-to-valley behavior of the histogram of double-quantized DCT coefficients, while the second approach is based on the distribution of the first digit of DCT coefficients. These methods have been extended to the study of the considered processing chain, for both the chain detection and the estimation of its parameters. More specifically, the proposed approaches provide an estimation of the quality factor of the previous JPEG compression and the amount of linear contrast enhancement.

Multiple Parenting Phylogeny Relationships in Digital Images

Recently, several studies have been concerned with modeling the parenthood relationships between near duplicates in a set of images. Two images share a parenthood relationship if one is obtained by applying transformations to the other. However, this is not the only form of parenting that can exist among images. An image might be a composition created through the combination of the semantic information existent in two or more source images, establishing a relationship between the sources and the composite. The problem of identifying these relations in a set containing near-duplicate subsets of source and composition images is referred to as multiple parenting phylogeny. Thus far, researchers tackled this problem with a three-step solution: 1) separation of near-duplicate groups; 2) classification of the relations between the groups; and 3) identification of the images used to create the original composition. In this work, we extend upon this framework by introducing key improvements, such as better identification of when two images share content, and improved ways to compare this content. In addition, we also introduce a new realistic professionally created data set of compositions involving multiple parenting relationships. The method we present in this paper is properly evaluated through quantitative metrics, established for assessing the accuracy in finding multiple parenting relationships. Finally, we discuss some particularities of the framework, such as the importance of an accurate reconstruction of phylogenies and the methods behavior when dealing with more complex compositions.

Multiple parenting identification in image phylogeny

Image phylogeny deals with tracing back parent-child relationships among near duplicates, images that share the same semantic content. This approach results in a visual structure showing the inheritance of semantic content among images, called phylogeny tree. In this paper, we extend upon the image phylogenys original formulation, which considers that an image may inherit content from only a single parent, to deal with situations whereby an image may inherit it from multiple different parents. Our objective is to find the multiple parenting relationships in a set of images, a problem which we refer to as multiple parenting phylogeny. The proposed solution works by first identifying near-duplicate groups and reconstructing their phylogenies; then among the found groups we determine the one(s) representing the composition images; finally, we detect the parenting relations between those compositions and the source images used to create them.

Unsupervised fusion for forgery localization exploiting background information

When image authenticity verification has to be carried out without any knowledge about the possible processing undergone by the image under analysis, it is fundamental to rely on a multi-clue approach, that merges the information stemming from several complementary forensic tools. This paper introduces a fully automatic framework for fusing the maps created by a set of unsupervised forgery localization algorithms, indicating possible manipulated areas. The framework takes into account the forgery maps, their reliability and the compatibility among the different traces considered by the tools. The achieved localization map is then refined by exploiting image content, thus improving the performance of the proposed system with respect to state of the art approaches.

New dissimilarity measures for image phylogeny reconstruction

Image phylogeny is the problem of reconstructing the structure that represents the history of generation of semantically similar images (e.g., near-duplicate images). Typical image phylogeny approaches break the problem into two steps: (1) estimating the dissimilarity between each pair of images and (2) reconstructing the phylogeny structure. Given that the dissimilarity calculation directly impacts the phylogeny reconstruction, in this paper, we propose new approaches to the standard formulation of the dissimilarity measure employed in image phylogeny, aiming at improving the reconstruction of the tree structure that represents the generational relationships between semantically similar images. These new formulations exploit a different method of color adjustment, local gradients to estimate pixel differences and mutual information as a similarity measure. The results obtained with the proposed formulation remarkably outperform the existing counterparts in the literature, allowing a much better analysis of the kinship relationships in a set of images, allowing for more accurate deployment of phylogeny solutions to tackle traitor tracing, copyright enforcement and digital forensics problems.

Identification of pictorial materials by means of optimized multispectral reflectance image processing

Image spectroscopy may allow identifying the materials present on a painting surface in a non-invasive way. The proposed method aims at optimizing, and thus reducing, the number of filters employed, while still providing a robust method, that achieves similar performances as traditional ones, which in turn employ a large number of filters. Moreover, we targeted the identification of the pigments present on the outer layer of a painting independently from their thickness, the underlying background or support, the binder employed, their aging and acquisition set-up. In order to achieve this objective, a relevant number of swatches have been prepared, on different supports and with different thicknesses and binding mediums. Spectral reflectance curves of such chemically known pictorial layers have been recorded by means of a spectrometer and a spectrophotometer. A novel Principal Component Analysis (PCA) based approach has been devised to select the most relevant wavebands, i.e. those that allow the most effective discrimination among (quasi)metameric colours, which are thus not to be distinguished with the naked eye or with an RGB camera. Comparisons of results using the 13 filters available on the filter wheel and of a selection of only 3 and 4 filters, support the idea of the simplified version investigated in this paper being a viable alternative.

Multispectral imaging for early diagnosis of melanoma

Melanoma is a very aggressive cutaneous neoplasm, incidence and mortality of which continues to rise worldwide. Identification of initial melanoma may be difficult because it may be clinically, and sometimes also dermoscopically, indistinguishable from benign lesions. Currently definitive diagnosis is made only by histopathological observation of the excised lesion. Several tools have been developed to help detecting malignant lesions. Dermoscopy highlights numerous characteristic features of the lesion and of the pigmented network. The method we propose exploits a multispectral imaging device to acquire a set of images in the visible and NIR range. Thanks to the fact that light propagates into the skin and reaches different depths depending on its wavelength, such a system is capable of imaging layers of structures placed at increasing depths. Therefore a new semeiotics is proposed to describe the content of such images. Dermoscopic criteria can be easily applied to describe each image in the set, however inter-images correlation needs new suitable descriptors. The first group of new parameters describes how the dermoscopic ones, vary across the set of images. More features are then introduced. E.g. the longest wavelength where structures can be detected gives an estimate of the maximum depth reached by the pigmented lesion. While the presence of a bright-to-dark transition between the wavebands in the violet to blue range, reveals the presence of blue-whitish veil, which is a further malignancy marker.

Noninvasive inspection of skin lesions via multispectral imaging

An optical noninvasive inspection tool is presented to, in vivo, better characterize biological tissues such as human skin. The method proposed exploits a multispectral imaging device to acquire a set of images in the visible and NIR range. This kind of information can be very helpful to improve early diagnosis of melanoma, a very aggressive cutaneous neoplasm, incidence and mortality of which continues to rise worldwide. Currently, noninvasive methods (i.e. dermoscopy) have improved melanoma detection, but the definitive diagnosis is still achieved only by invasive method (istopathological observation of the excised lesion). The multispectral system we developed is capable of imaging layers of structures placed at increasing depth, thanks to the fact that light propagates into the skin and reaches different depths depending on its wavelength. This allows to image many features which are less or not visible in the clinical and dermoscopic examination. A new semeiotics is proposed to describe the content of multispectral images. Dermoscopic criteria can be easily applied to describe each image in the set, however inter-images correlations need new suitable descriptors. The first group of new parameters describes how the dermoscopic features, vary across the set of images. More aspects are then introduced. E.g. the longest wavelength where structures can be detected gives an estimate of the maximum depth reached by the pigmented lesion. While the presence of a bright-to-dark transition between the wavebands in the violet to blue range, reveals the presence of blue-whitish veil, which is a further malignancy marker.

Improving on fast and automatic texture mapping of 3D dense models

Not all range devices acquire, along with 3D data, the objects texture. Moreover, not always the desired texture is the visible light image. In such cases, currently, an “a posteriori” texturing of a 3D model is mostly performed in a manual or semi-automated fashion, resulting in a subjective and time consuming operation. Matching homologues points between 2D and 3D data in fact proved to be a more complex operation than image to image, or geometry to geometry registration. The method described in this paper is designed to be fully automated. The software takes as input a generic un-textured 3D model and a nonspecific texture image, which could be different from a visible light photograph, but belong to a set of diagnostic images like X rays, UV light, or IR images. It relies on the creation from the 3D model of several 2D depth maps which retains an exact correspondence with the points of the relief. Each depth map is generated from a different external “view point”. The number and location of such viewpoints is determined “a priori”, but their final position is to be changed and adjusted on a iterative and automatic base, to assure the possibility of an optimal choice. The selection of the best matching depth map is done by picking the depth map which shows the highest similarity with the texture image, based on a 2D-2D registration procedure performed on all generated depth maps. In order to speed up the procedure, a multi-resolution approach is adopted, where the coarse selection is performed on down-sampled images. Cross correlation and Maximization of Mutual Information (MMI) are here both used as similarity measures, exploiting their different and complementary performances depending on the image size.