Tian-Tsong Ng

Blind detection of photomontage using higher order statistics

We investigate the prospect of using bicoherence features for blind image splicing detection. Image splicing is an essential operation for digital photomontaging, which in turn is a technique for creating image forgery. We examine the properties of bicoherence features on a data set, which contains image blocks of diverse image properties. We then demonstrate the limitation of the baseline bicoherence features for image splicing detection. Our investigation has led to two suggestions for improving the performance of bicoherence features, i.e., estimating the bicoherence features of the authentic counterpart and incorporating features that characterize the variance of the feature performance. The features derived from the suggestions are evaluated with support vector machine (SVM) classification and is shown to improve the image splicing detection accuracy from 62% to about 70%.

Physics-motivated features for distinguishing photographic images and computer graphics

The increasing photorealism for computer graphics has made computer graphics a convincing form of image forgery. Therefore, classifying photographic images and photorealistic computer graphics has become an important problem for image forgery detection. In this paper, we propose a new geometry-based image model, motivated by the physical image generation process, to tackle the above-mentioned problem. The proposed model reveals certain physical differences between the two image categories, such as the gamma correction in photographic images and the sharp structures in computer graphics. For the problem of image forgery detection, we propose two levels of image authenticity definition, i.e., imaging-process authenticity and scene authenticity, and analyze our technique against these definitions. Such definition is important for making the concept of image authenticity computable. Apart from offering physical insights, our technique with a classification accuracy of 83.5% outperforms those in the prior work, i.e., wavelet features at 80.3% and cartoon features at 71.0%. We also consider a recapturing attack scenario and propose a counter-attack measure. In addition, we constructed a publicly available benchmark dataset with images of diverse content and computer graphics of high photorealism.

A model for image splicing

The ease of creating image forgery using image-splicing techniques will soon make our naive trust on image authenticity a tiling of the past. In prior work, we observed the capability of the bicoherence magnitude and phase features for image splicing detection. To bridge the gap between empirical observations and theoretical justifications, in this paper, an image-splicing model based on the idea of bipolar signal perturbation is proposed and studied. A theoretical analysis of the model leads to propositions and predictions consistent with the empirical observations.

Passive-blind Image Forensics

Publisher Summary This chapter discusses passive-blind image forensics (PBIF). PBIF is concerned with two problems: image forgery detection and image source identification. Most of the image forgery detection techniques are associated to the specific image forgery creation techniques. Forgery detectors uncover act of fabrication by assessing the authenticity of a given image. There are two approaches for image forgery detection: detecting the authentic characteristics of images and detecting the telltale characteristics specific to the image forgery creation techniques. The goal of the passive-blind image source identification is to identify the type of image source. Identification of the image source helps in deciding whether an image is acceptable for a specific application. One problem of concern in image source identification is the classification of photographic images (PIMs) and photorealistic computer graphics (PRCGs), so the chapter summarizes the differences between PIMs and PRCGs––namely, object model difference, light transport difference, and acquisition difference. The chapter gives a reference of data set by enlisting the category of images in it and concludes with a review of the automatic or the semi-automatic computer techniques for image forgery creation.

Using Geometry Invariants for Camera Response Function Estimation

In this paper, we present a new single-image camera response function (CRF) estimation method using geometry invariants (GI). We derive mathematical properties and geometric interpretation for GI, which lend insight to addressing various algorithm implementation issues in a principled way. In contrast to the previous single-image CRF estimation methods, our method provides a constraint equation for selecting the potential target data points. Comparing to the prior work, our experiment is conducted over more extensive data and our method is flexible in that its estimation accuracy and stability can be improved whenever more than one image is available. The geometry invariance theory is novel and may be of wide interest.

Camera response function signature for digital forensics - Part I: Theory and data selection

Camera response function (CRF) is a form of camera signatures which can be extracted from a single image and provides a natural basis for image forensics. CRF extraction from a single-image is in theory ill-posed. It relies on specific structures in an image that offer glimpses of the CRF. Therefore, the challenges in CRF extraction are first in identifying structures of such property, second in locating such structures in an image, and third in extracting the CRF attributes from the selected structures. In our past work, we proposed that CRF attributes can be found on linear structures in an image and extracted using linear geometric invariants. In this paper, we show additional properties on linear geometric invariants, propose a more robust way to select linear structures in an image, and provide a model-based method to extract CRF attributes from the linear structures. This paper is divided into two parts. Part I is devoted to the theory of linear geometric invariants and the robust selection of linear structures. The linear structure candidates obtained from the method in Part I are used to instantiate the edge profiles for CRF extraction in Part II. The paper as a whole presents a reliable method for CRF extraction, together with rigorous analysis which gives useful insights into the method. In the first half of Part I, a simpler proof that links the equality of linear geometric invariants to a linear-isophote surface is given. As a by-product, the proof leads to an additional way to detect linear-isophote surfaces which uses only the first-order partial derivatives and improves detection reliability. In the second half of Part I, the variance of linear geometric invariants is shown to have a structure which can be used to improve the robustness in detecting linear-isophote surfaces.

An online system for classifying computer graphics images from natural photographs

We describe an online system for classifying computer generated images and camera-captured photographic images, as part of our effort in building a complete passive-blind system for image tampering detection (project website at http: //www.ee.columbia.edu/trustfoto). Users are able to submit any image from a local or an online source to the system and get classification results with confidence scores. Our system has implemented three different algorithms from the state of the art based on the geometry, the wavelet, and the cartoon features. We describe the important algorithmic issues involved for achieving satisfactory performances in both speed and accuracy as well as the capability to handle diverse types of input images. We studied the effects of image size reduction on classification accuracy and speed, and found different size reduction methods worked best for different classification methods. In addition, we incorporated machine learning techniques, such as fusion and subclass-based bagging, in order to counter the effect of performance degradation caused by image size reduction. With all these improvements, we are able to speed up the classification speed by more than two times while keeping the classification accuracy almost intact at about 82%.

Predicting optimal operation of MC-3DSBC multidimensional scalable video coding using subjective quality measurement

Recently we have witnessed a growing interest in the development of the subband/wavelet coding (SBC) technology, partly due to the superior scalability of SBC. Scalable coding provides great synergy with the universal media access applications, where media content is delivered to client devices of diverse types through heterogeneous channels. In this respect, SBC system provides flexibility in realizing different ways of media scaling, including scaling dimensions of SNR, spatial, and temporal. However, the selection of specific scalability operations given the bit rate constraint has always been ad hoc - a systematic methodology is missing. In this paper, we address this open issue by applying our content-based optimal scalability selection framework and adopting subjective quality evaluation. For this purpose we firstly explore the behavior of SNR-Spatial-Temporal scalability using Motion Compensated (MC) SBC systems. Based on the system behavior, we propose an efficient method for the optimal selection of scalability operator through content-based prediction. Our experiment results demonstrate that the proposed method can efficiently predict the optimal scalability operation with an excellent accuracy.

Discrimination of Computer Synthesized or Recaptured Images from Real Images

An image that appears to be a photograph may not necessarily a normal photograph as we know it. For example, a photograph-like image can be rendered by computer graphics instead of being taken by a camera or it can be a photograph of an image instead of a direct photograph of a natural scene. What is really different between these photographic appearances is their underlying synthesis processes. Not being able to distinguish these images poses real social risks, as it becomes harder to refute claims of child pornography as non-photograph in the court of law and easier for attackers to mount an image or video replay attack on biometric security systems. This motivates digital image forensics research on distinguishing these photograph-like images from true photographs. In this chapter, we present the challenges, technical approaches, system design and other practical issues in tackling this multimedia forensics problem. We will also share a list of open resources and the potential future research directions in this area of research which we hope readers will find useful.

Identifying and prefiltering images

Given the ability of photorealistic computer graphics (photorealistic CG) to emulate photographic images, as seen in movies and the print media today, there is little doubt that uninformed viewers can easily mistake photorealistic computer- generated graphics for photographic images. In fact, there was already evidence 20 years ago that to the naked eye certain computer graphics were visually indistinguishable from photographic images. Such convincing photorealism qualifies computer graphics as a form of image forgery that can be unscrupulously exploited. Some popular Web sites even highlight examples of computer generated photorealism that human eyes find indistinguishable from photographic images.