The possibility of uncovering the sequence of pressed keys in a PIN-pad, ATM or other device with mechanical keyboard that processes sensitive information is a serious security threat. In this paper, we show that the Acoustic Transfer Function (ATF) of the device carries information that discloses the pressed key. To demonstrate this hypothesis, we glued two microphones in empty spaces of two different PIN-pads, collected the acoustic emission generated by the keystrokes and estimated the transfer functions of the system through the two acquired audio signals. A simple classification scheme using the ATFs yielded 99.6% of recognition accuracy in one device and 46% in the other. The results show that there are devices that can be easily attacked using ATF and those that are more difficult to attack using this technique. We analyzed again the data where we obtained 99.6% of accuracy, this time using the traditional frequency spectrum based technique, obtaining only 52% of recognition rate. The results also show that the certification process has failed in identifying this vulnerability.

This paper describes an attack that identifies the sequence of keystrokes analyzing mechanical vibrations generated by the act of pressing keys. We use accelerometers as vibration sensors. The apparatus necessary for this attack is inexpensive and can be unobtrusively embedded within the target equipment. We tested the proposed attack on an ATM keypad and a PIN-pad. We achieved the key recognition rates of 98.4% in ATM keypad, 76.7% in PIN-pad resting on a hard surface, and 82.1% in PIN-pad hold in hand.

A near-planar object seen from different viewpoints results in differently deformed images. Under some assumptions, viewpoint changes can be modeled by affine transformations. Shape features that are affine-invariant (af-in) must remain constant with the changes of the viewpoint. Similarly, shape similarity metrics that are af-in must rate the difference between two shapes, regardless of their viewpoints. Af-in shape features and similarity metrics can be used for the shape classification and retrieval. In this paper, we propose a new set of af-in shape features and similarity metrics. They are based on the area matrix, a structure that contains multiscale information about the shape. Experimental results indicate that the proposed techniques are robust to viewpoint changes and can rate correctly the dissimilarities between the shapes.

An authentication watermarking technique (AWT) inserts some hidden data into an image in order to detect any accidental or malicious image alteration. Recent papers have proposed some cryptography-based AWTs for binary and halftone images, but none of them seems to be well-suited for public-key authentication of binary document images. Many of the previous public-key AWTs for binary images can be assaulted by "parity attacks." This paper proposes a new AWT for binary document images that can detect any modification of the watermarked document, even a single pixel flipping. Moreover, both its secretand public-key versions are completely immune against parity attacks. Document images watermarked by the proposed technique present high visual quality, without salt-and-pepper noise.

This paper presents a new, simple and elegant technique to improve the detection of brain regions with increased neuronal activity in functional magnetic resonance imaging (fMRI). This technique is based on the robust anisotropic diffusion (RAD). A direct application of RAD to fMRI does not work, mainly due to the lack of sharp boundaries between activated and non-activated regions. To overcome this difficulty, we propose to estimate the statistical parametric map (SPM) from the noisy fMRI, compute the diffusion coefficients in the SPM-space, and then perform the diffusion in the structural information-removed fMRI data using the coefficients previously computed. These steps are iterated until the convergence. We have tested the new technique in both simulated and real fMRI, obtaining surprisingly sharp and noiseless SPMs with increased statistical significance. We use receiver operating characteristics (ROC) curves to show that the proposed technique is superior than the conventional correlation method.

An authentication watermarking technique (AWT) inserts hidden data into an image in order to detect any accidental or malicious alteration to the image. AWT usually computes, using a secret-key, the message authentication code (MAC) of the image, and somehow inserts MAC into the image itself. The authentication verification is performed using the same secret-key used in the insertion. JBIG2 is an international standard for lossy and lossless binary image compression. It decomposes the image into several regions (text, halftone, line-art) and encodes each one using the most appropriate method. This paper proposes a novel data-hiding technique to embed the data into the text region of a JBIG2 file. We, then, use it to create an AWT for a JBIG2-encoded image. The embedded data can be extracted from either the JBIG2 file itself or the binary image obtained by decoding the JBIG2 file.

An object seen from different viewpoints results in differently deformed images. Affine-invariant shape classification must classify correctly the object, regardless its viewpoint. In this paper, we propose new local and global features invariant under affine transformation. These features can be used for supervised or unsupervised shape classification, and for shape-based image indexing and retrieval. One of the proposed features is related to the convex deficiency and the others are extracted from the area matrix. Area matrix was used by Shen for the similarity matching in image retrieval. However, differently from the Shen's work, we parameterize the shape contour using the affine-length parameter. This makes our features robust to affine parameterization, while Shen's work does not have this property. Experimental results indicate that our method can classify correctly even highly deformed and noisy shapes using small training sets.

Authentication watermarking is the process that inserts hidden data into an object (image) in order to detect any fraudulent alteration perpetrated by a malicious hacker. In the literature, quite a small number of secure authentication methods are available for binary images. This paper proposes a new secure authentication watermarking method for binary images. It can detect any visually significant alteration while maintaining good visual quality. As usual, the security of the algorithm lies on the secrecy of a private-key. Only its owner can insert the correct watermark while anyone may verify the authenticity through the corresponding public-key. A possible application of the proposed technique is in Internet fax transmission, i.e. for legal authentication of documents routed outside the phone network.

This paper presents a new, accurate, and efficient technique to increase the spatial resolution of binary halftone images. It makes use of a machine learning process to automatically design a zoom operator starting from pairs of input-output sample images. To accurately zoom a halftone image, a large window and large sample images are required. Unfortunately, in this case, the execution time required by most of the previous techniques may be prohibitive. The new solution overcomes this difficulty by using decision tree (DT) learning. Original DT learning is modified to obtain a more efficient technique (WZDT learning). It is useful to know, a priori, sample complexity (the number of training samples needed to obtain, with probability 1 - delta, an operator with accuracy epsilon): we use the probably approximately correct (PAC) learning theory to compute the sample complexity. Since the PAC theory usually yields an overestimated sample complexity, statistical estimation is used to evaluate, a posteriori, a tight error bound. Statistical estimation is also used to choose an appropriate window and to show that DT learning has good inductive bias. The new technique is more accurate than a zooming method based on simple inverse halftoning techniques. The quality of the proposed solution is very close to the theoretical optimal obtainable quality for a neighborhood-based zooming process using the Hamming distance to quantify the error.

Inverse halftoning is the process to retrieve a (gray) continuous-tone image from a halftone. Recently, machine-learning-based inverse halftoning techniques have been proposed. Decision-tree learning has been applied with success to various machine-learning applications for quite some time. In this paper, we propose to use decision-tree learning to solve the inverse halftoning problem. This allows us to reuse a number of algorithms already developed. Especially, the maximization of entropy gain is a powerful idea that makes the learning algorithm to automatically select the ideal window as the decision-tree is constructed. The new technique has generated gray images with PSNR numbers, which are several dB above those previously reported in the literature. Moreover, it possesses very fast implementation, lending itself useful for real time applications.

We present a new image reconstruction technique from acquisition data in divergent geometry distributed over a limited angle. In many practical situations, it is not possible to obtain the data distributed over the whole 360/spl deg/. For example, when the test object presents anisotropic properties, i.e., when it is extremely sensible to the radiation coming from some specific direction. Physical constraints can also obstruct the acquisition of data in all angles. The proposed technique solves this problem by allowing image reconstruction from data distributed over a limited angle. This technique first converts the divergent projections distributed over a limited angle into parallel projections homogeneously distributed over the whole contour of the test object. Then, an algorithm based on extended entropy reconstructs the image.

In this paper, we describe some weaknesses of public-key blockwise fragile authentication watermarkings and the means to make them secure. Wong's (1998) original algorithm is not secure against a mere block cut-and-paste or the well-known birthday attack. To make it secure, some schemes have been proposed to make the signature of each block depend on the contents of its neighboring blocks. We attempt to maximize the change localization resolution using only one dependency per block with a scheme we call hash block chaining version 1 (HBC1). We then show that HBC1, as well as any neighbor-dependent scheme, are susceptible to another forgery technique that we have named a transplantation attack. We also show a new kind of birthday attack that can be effectively mounted against HBC1. To thwart these attacks, we propose using a nondeterministic digital signature together with a signature dependent scheme (HBC2). Finally, we discuss the advantages of using discrete logarithm signatures instead of RSA for watermarking.

Digital halftone is the technique used to convert gray-scale images into binary ones, simulating gray shades by scattering appropriately black and white pixels. Sometimes, there arises the necessity of increasing the resolution of a halftone image. Some recent works have proposed a number of learning-based techniques to zoom binary images. However, they cannot consider a large neighborhood to decide the colors of the resolution-increased pixels, because their running time skyrockets with the growth of the window and the sample images sizes. The use of large window and samples is required to zoom halftone images accurately. This paper presents a new technique to zoom quickly and precisely images generated by any locally-decided halftone algorithm. It is based on decision-tree learning and it is very fast, even using a large window or large samples. The zoomed images obtained by this technique are incredibly sharp and accurate.

In a typical office environment, heterogeneous devices and software, each working in a different spatial resolution, must interact together. Thus, there frequently arises the resolution conversion problem. The present paper addresses the spatial resolution increasing of binary images and documents (for example, conversion of a 300 dpi image into 600 dpi). A new and efficient solution to this problem is proposed. It makes use of k-nearest neighbor learning to automatically design the optimal windowed zoom operator starting from pairs of in-out sample images. The k-nearest neighbor learning has good inductive bias that allows reducing the quantity of the training sample images needed. The resulting operator is stored in a look-up-table, which is extremely fast computationally.

Nonlinear filters have been used not only in noise elimination but in a wide variety of image processing applications. Traditionally the design of a digital filter is a manual task and the user accomplishes it based on previous experiences. Unfortunately often this is not a trivial task. Thus some works try to overcome this difficulty constructing the filters automatically by computational learning neural networks, genetic algorithms and statistical estimation. These works use typical input-output images of the application as the training samples. Many different kinds of filters can be easily constructed using this approach. This paper proposes the use of nearest neighbor (NN) learning to automatic filter construction. The kd-tree (k-dimensional binary tree) is used to accelerate the NN searching. A texture recognition application example is depicted.