With digital images replacing their analog counterparts in more and more venues, and increasing functionality of image editing softwares, reliable forensic tools that help establish the origin, authenticity, and the chain of custody of digital images are becoming indispensable. These tools can prove to be vital whenever questions of digital image integrity are raised. There are various levels at which the image source identification problem can be addressed. One may want to find the particular device (digital camera or scanner), which generated the image, or one might be interested in knowing only the make and model of the device. We have developed reliable methods for image source classification and source scanner identification. Dividing images into camera generated or scanned images, is an essential step before applying the source identification algorithms specific to one class or the other. Source scanner identification may be of crucial importance in situations such as e-commerce between banks using scanned checks. Two key steps of our image forensics algorithms are extraction of appropriate features from the images and training of classifiers such as SVM. Our feature vector based methods allow close to 90% classification accuracy on a set of ten scanners and ten digital cameras. Future work in this direction will allow reliable image source identification from images undergone post-processing such as cropping, compression, sub-sampling, rotation and different kinds of filtering operations.

This project involves the development of cryptographic protocols and techniques to enhance the security and privacy of the collection, storage, and processing of online survey data. Specifically, cryptographic protocols are designed for carrying out computations on online survey data in such a manner that only the approved output is revealed by the protocol, not the input data. The protocol therefore acts as a purpose-enforcement mechanism such that the data can be used only as authorized and can no longer be collected for one purpose and illicitly used for another. This approach has the benefit of protecting collected online survey data from hackers, spyware, misbehaving insiders, and accidental disclosure through the loss of a storage device. The security and purpose-enforcement in turn make it possible to potentially acquire higher quality responses to surveys on sensitive topics, a significant benefit in view of the increasing societal reliance on surveys from which important policy conclusions are often drawn in the health sciences, social and behavioral sciences, experimental economics, and other disciplines where the human subjects’ individual responses are to be considered sensitive and kept private. The cryptographic protocols are tested through the administration of an online survey on Internet addiction. Protocols for data analysis techniques are developed to aid in the analysis of the collected data. The students involved in this project acquire a unique combination of inter-disciplinary talents. The results of this work are disseminated through the project Web site: http://www.cs.purdue.edu/homes/mja/secsurv/SecureSurv.html

In 1953, Claude Shannon, the founder of information theory, pointed out that there is no theory via which information embodied in structure can be quantified; this situation remains in effect today. The need for such a theory has become pressing in recent years with the proliferation of structured data sets arising from diverse applications. We have yet to answer fundamental questions such as:

What are fundamental limits on storage and processing of structural information? What are fundamental bounds on extraction of information from large biological databases? Lack of understanding of such questions threatens to raise severe impediments to further advances in science and engineering of complex systems. The main goal of this work is to search for measures and algorithms to appraise the amount of organization and structure embodied in artifacts and natural objects.

We propose to make headway in information theory of data structures.

Data is increasingly available in various forms (e.g., sequences, expressions, interactions, structures) and in exponentially increasing amounts. Most of such data is multidimensional and context dependent; thus it necessitates novel theory and efficient algorithms to extract meaningful information from non-conventional data structures. In compressing such a data structure, one must take into account two types of information: the information conveyed by the structure itself, and then the information conveyed by the data labels implanted in the structure. The specific goals of this project are: (i) characterization of the total amount of information conveyed by a data structure (and how this decomposes into the two types of information mentioned above), and (ii) the design of efficient compression algorithms based upon the total amount of information conveyed in (i).

This paper presents a novel invisible robust watermarking scheme for embedding and extracting a digital watermark in an image. The novelty lies in determining a perceptually important sub-image in the host image. Invisible insertion of the watermark is performed in the most significant region of the host image such that tampering of that portion with an intention to remove or destroy will degrade the esthetic quality and value of the image. One feature of the algorithm is that this sub-image is used as a region of interest for the watermarking process and eliminates the chance of watermark removal. Another feature of the algorithm is the creation of a compound watermark using the input user watermark (logo) and attributes of the host image. This facilitates the homogeneous fusion of a watermark with the cover image, preserves the quality of the host image, and allows robust insertion-extraction. Watermark creation consists of two distinct phases. During the first phase, a statistical image is synthesized from a perceptually important sub-image of the image. A compound watermark is created by embedding a watermark (logo) into the statistical synthetic image by using a visible watermarking technique. This compound watermark is invisibly embedded into the important block of the host image. The authentication process involves extraction of the perceptive logo as well statistical testing for two-layer evidence. Results using standard benchmarks demonstrates the robustness and efficacy of the proposed watermarking approach. Ownership proof could be established under various hostile attacks.