This PDF file contains the front matter associated with SPIE Proceedings Volume 7075, including the Title Page, Copyright information, Table of Contents, Introduction, and the Conference Committee listing.
Data compression ideas can be extended to assess the data quality across multiple sensors to manage the network of sensors to optimize the location accuracy subject to communication constraints. From an unconstrained-resources viewpoint it is desirable to use the complete set of deployed sensors; however, that generally results in an excessive data volume. We have previously presented here results on selecting pre-paired sensors. We have now extended our results to enable optimal joint pairing/selection of sensors.
Pairing and selecting sensors to participate in sensing is crucial to satisfying trade-offs between accuracy and time-line requirements. We propose two methods that use Fisher information to determine sensor pairing/selection. The first method optimally determines pairings as well as selections of pairs but with the constraint that no sensors are shared between pairs. The second method allows sensors to be shared between pairs. In the first method, it is simple to evaluate the Fisher information but is challenging to make the optimal selections of sensors. However, the opposite is true in the second method: it is more challenging to evaluate the Fisher information but is simple to make the optimal selections of sensors.
Early work in source location using time-difference-of-arrival/frequency-difference-of-arrival (TDOA/FDOA) focused on locating acoustic sources while later work focused on locating electromagnetic sources. The key difference is the signal model assumptions: WSS Gaussian process is widely used in the acoustic case but is not appropriate in the electromagnetic case. The Fisher information (FI) is fundamentally different for the two scenarios and leads to different distortion metrics for data compression algorithms that seek to maximize the FI for a given data rate. We discuss the philosophical impacts of this relevant to the following question: having collected a single set of data and wanting to do the best "job" for that data, should it matter if the data is viewed as coming from a WSS random process?
This work shows that one must be careful when using a random signal model. If one takes the operational rate-distortion view, the goal of compression is to adapt the algorithm to the specific data observed. This is a modern view that contrasts with classical rate-distortion where the distortion measure includes an averaging over the ensemble. We assert that for the operational rate-distortion approach with FI as distortion measure, one should not use a random signal model.
We examine the effects of wavelet compression on target detection algorithms when the targets are single-pixel point
sources modulated by the point-spread of an optical system. The experimental data combines frames collected from a
multispectral sensor with simulated targets based on an Airy function. We studied several different types of wavelets
and found that the Daubechies 2 wavelet resulted in the best overall target detection and fewest false alarms with
increasing compression. Results show that wavelet compression may decrease pixel intensities, increase target signal-to-noise
ratio, and reduce false detections. Consequently it may negatively affect target detection unless the detector is
designed to take the decreased intensity into account.
In this paper, we present an approach to predict perceived quality of compressed images while incorporating real visual attention coordinates. Information about the visual attention is not usually taken into account in image quality assessment models. The idea of implementing gaze information into the image quality assessment system lies in that the artefacts are more disturbing to human observer in the region with higher saliency than in other parts of an image. Impact of the re¬gion of interest on estimation accuracy of a simple image quality metric is investigated. The gaze coordinates were calculated using calibrated electro-oculogram records of human observers while watching a number of test images. The same images were used for subjective image quality assessment. Obtained mean opinion scores of perceived image quality and eye tracking data were used to verify potential improvement of assessment accuracy for a simple image quality metric. Based on the proven effect, our previously developed system for still image quality assessment has been adapted while utilizing information about the visual attention. The potential performance improvement of existing image coding while incorporating the spatially adaptive HVS is discussed.
Due to the huge amount of data and the increasing usage, compression of 4D medical data sets is essential. These datasets
consist of a number of sampled volume elements varying in time and are compressed either with spatial transformation
based (e.g. JPEG2000-3D) or motion estimation based schemes. This paper presents a combined approach incorporating
both, temporal and spatial information at the same time to compress 4D medical datasets. It is adopting a very similar
four-dimensional Multi-View-Coding (MVC) scheme which is known from video processing to 4D medical datasets and
compares experimental results with H.264 compression. Rate distortion characteristics show the advantages of such a
combined spatio-temporal approach.
Images contain large amount of data and are used in many applications.
Not only compressed image data save space, but also in certain applications
such as the World-Wide-Web (WWW),
they save time since the amount of data to be transmitted is smaller than the original image.
Codebook approaches have been used in lossy image compression.
For example, vector quantization employs codebook.
For lossless compression of images, more than a decade ago,
a codebook approach based on spanning trees was proposed.
In this work we investigate suitability of permutation codes for still images.
High speed, low complexity, and interoperability are just three of the main advantages turning the MPEG stream
watermarking into a hot research topic. Unfortunately, viable solutions (in terms of robustness, data payload and
transparency) are yet to be found. In their previous work, the authors computed general models for the watermarking
attack effects (StirMark, linear & nonlinear filtering, rotations) in the MPEG-4 AVC stream. These models (expressed as
noise matrices) are now the starting point for evaluating three classes of watermarking insertion techniques (substitutive,
additive, and multiplicative). For each class, a specific set of noise matrices is first computed by particularising the
general model. Secondly, the corresponding capacity values (i.e. the largest data payload which can be inserted for
prescribed transparency and robustness) are computed. The paper is concluded with a comparison among these method
performances. The experiments are run on a video corpus of 10 video sequences of about 25 minutes each.
In this work, we extend our previous research on gray level co-occurrence matrix (GLCM) based watermark embedding in the discrete cosine transform (DCT) domain to the discrete wavelet transform (DWT) domain. The GLCM method incorporated human visual system information into the embedding process making the watermark more transparent. DWT techniques allow for more compression as fewer coefficients are required to reconstruct an image. In addition, DWT methods will not exhibit block artifacts commonly encountered when applying block based DCT methods. The watermark identification is further enhanced using neural networks. In this research, daubechies wavelets are utilized to evaluate the efficiency of the watermark identification while the method is subjected to multiple attacks such as filtering, compression, or rotation. The results are then compared with previously published methods by the authors such as LMS based correlation and adaptive DWT based watermark identification.
In previous research, we have shown the ability of neural networks to improve the performance of the watermark system to identify the watermark under different attacks. On the other hand, in this work we apply neural networks to embed the watermark in the discrete wavelet transform (DWT) domain. We then use features based on principal component analysis (PCA) to blindly identify the watermark. PCA reduces the dimensionality as well as the redundancies of the data. Neural networks classifiers are implemented to determine whether the watermark is present. Different features are used to test the performance of the method. The efficacy of the technique is then compared to previous techniques such as the gray level co-occurrence matrix (GLCM) based or the LMS enhanced watermark identification. The comparative results from the previously used methods are presented in this paper.
Nowadays, nobody doubts about the huge economical benefits the watermarking solutions will one day bring. The paper
is devoted to the theoretical evaluation of the watermarking capacity, i.e. devoted to find out with mathematical rigour
the maximum amount of information which can be inserted into the DWT of natural video, for prescribed constraints of
transparency and robustness. The starting point is the accurate statistical model for the watermarking attacks the authors
already reported. In this paper, in addition to the classical Shannon solutions, the capacity is evaluated by two
approaches: (1) a method developed in order to increase speed and precision for watermarking evaluations and (2) the
general Blahut-Arimoto algorithm, adapted by Justin Dauwels for the discrete case. The experiments are run on a video
corpus of 10 video sequences of about 25 minutes each.
Digital watermarking of images, the act of hiding a message inside an image, is still a young, yet growing, research
field. We have developed an environment in Matlab that allows researchers, teachers and students alike to get
acquainted with the concepts of digital image watermarking techniques. This user-friendly environment is divided
into two parts. First there is the educational part, which visualizes watermarking techniques and gives users the
possibility to observe the results of various attacks. This part can be used as a demonstrator during lectures
on image watermarking. The second part is dedicated to a benchmarking tutorial section, which allows users to
easily compare watermarking techniques. A user new to the field of benchmarking can simply insert existing or
newly developed watermarking algorithms, attacks and metrics into the benchmarking tutorial environment. We
also included an attack section that is easy to adjust and extend with new attacks. Furthermore, we provided a
report generator that will summarize all the benchmarking results in a clear, visually appealing manner. Through
the use of Matlab, we were able to make a framework that is easy to modify, update and expand and also enables
everyone to use the existing signal processing libraries of Matlab.
In this paper, we establish the optimum interpolation approximation for a set of multi-dimensional statistical
orthogonal expansions. Each signal has a bounded linear combination of higher order self-correlations and
mutual-correlations with respect to coefficients of the expansion. For this set of signals, we present the optimum
interpolation approximation that minimizes various worst-case measures of mean-square error among all the linear
and the nonlinear approximations. Finally, as a practical application of the optimum interpolation approximation,
we present a discrete numerical solution of linear partial differential equations with two independent variables.
We present the optimum running-type approximation of FIR filter bank that minimizes various worst-case measures of error, simultaneously, with respect to each of two different sets of signals. The first set is a set of piecewise analytic time-limited but approximately band-limited signals. When a supreme signal that realizes the prescribed worst-case measure of error exists, we prove, firstly, that there exists one-to-one correspondence between error in a wide time interval and error in a small interval. Based on this one-to-one correspondence, we prove that the approximation presented in this paper is the optimum in some sense. Secondly, we consider a set of band-limited signals with a main-lobe and a pair of small side-lobes and obtain similar conclusion.
There are certain classes of astronomical objects that have rather involved spectra that can also be a composite of a number of different spectral signatures, as well as spatial characteristics that can be used for identification and analysis. Such objects include galaxies and quasars with active nuclei, colliding / interacting galaxies, and globular cluster systems around our own Milky Way and other galaxies. Flash hyperspectral imaging adds coherence-time limited functionality so that Earth orbiting spacecraft and solar system objects such as planets, asteroids and comets can be spectrally imaged as well, as these also have both spatial and spectral structure rotating and moving within much shorter time spans. Flash hyperspectral imaging systems are, therefore, also useful for faster simultaneous spatial and spectral feature analysis. Previous work has explored spectral unmixing and other types of feature extraction of these general types of objects, but without consideration of the hyperspectral imaging system involved, neither in how the data is collected nor in how the datacube is reconstructed. We will present a proof of concept simulation of a resolved object as it is imaged through such a physically modeled imaging system and its datacube reconstructed. Finally, we provide a demonstration of the capability with astronomical data, Venus and a binary star, when constrained by our physical model of the instrumental transfer function.
A novel approach is presented for edge detection using the USAN area feature at a pixel, since the USAN area
characterizes the structure of the edge present in the neighborhood of a pixel. Next, Gaussian edge detector mask is
applied at each pixel taking the membership function of the USAN area. The resultant is then modified and thresholded
to yield the binary image. The results of the proposed edge detector is compared with other well known edge detector
like Canny, SUSAN, Gradient diffusion operator etc. The edge detected image from the proposed approach seems to fare
well over others. Moreover this edge detector also preserves the structure of the image unlike Canny. This is an added
advantage with this approach.
In the field of biology, compensation of multiple fluorescence is necessary for quantitative measurement of samples.
Conventional methods depend on a linear combination model irrespective of the model's adequacy. Therefore,
proper compensation has not been performed in some situations. To overcome this problem, we propose a
method for performing nonlinear mapping with emphasis on the distribution of samples on a plane of which the
base vectors are references of fluorescence. This paper describes an experiment with measurement of multiple
fluorescence and compensation using a conventional method and our proposed method. Results show that
multiple fluorescence is not always able to assume a linear combination model. Moreover, we confirmed that the
presented method is independent of linearity of multiple fluorescence.
Quality assessment for remote sensing image compression is of great significance in many practical applications. A
comprehensive index based on muti-dimensional structure model was designed for image compression assessment,
which consists of gray character distortion dimension, texture distortion dimension, loss of correlation dimension. Based
on this model, a new comprehensive image quality index-Q was proposed. In order to assess the agreement between our
comprehensive image quality index Q and human visual perception, we conducted subjective experiments in which
observers ranked reconstructed images according to perceived distortion. For comparison, PSNR is introduced. The
experiments showed that Q had a better consistency with subjective assessment results than conventional PSNR.
In this paper, new Gaussian mixture classifiers are designed to deal with the case of an unknown number of mixing
kernels. Not knowing the true number of mixing components is a major learning problem for a mixture classifier
using expectation-maximization (EM). To overcome this problem, the training algorithm uses a combination of
covariance constraints, dynamic pruning, splitting and merging of mixture kernels of the Gaussian mixture to
correctly automate the learning process. This structural learning of Gaussian mixtures is employed to model
and classify Hyperspectral imagery (HSI) data. The results from the HSI experiments suggested that this new
methodology is a potential alternative to the traditional mixture based modeling and classification using general
Many experiments used light scattering to visualize the fluctuations of fluid's density. Fluids near the critical point are
affected by gravity because the compressibility of the fluid is very large near the critical point. Therefore, microgravity
experiments allowed new phenomena to be discovered by reducing convection, sedimentation and buoyancy
In order to study, fluctuation and phase separation processes near the critical point of pure fluids without the influence of
the Earth's gravity, a number of experiments were performed in microgravity. Our results refer to a set of experiments
that studied local density fluctuations by illuminating a cylindrical cell filled with sulfur hexafluoride, near its liquid-gas
critical point. Using image analysis, we estimated the temperature of the fluid in microgravity from the recorded images
showing fluctuations of the transmitted and scattered light. Our method has the advantage of avoiding any reference to
the spatial correlation of the pixels in the recorded images. We assumed that the variation of the scattered light intensity
is proportional to the average value of the gray levels. Furthermore, we also assumed that a small fluctuation of the fluid
density induces a change in the scattered light intensity that can be measured from average gray scale intensity of the
image. We found that the histogram of an image can be fitted to a Gaussian relationship and by determining its width we
were able to estimate the position of the critical point.
The paper will introduce the quaternary, or radix 4, based system for use as a fundamental standard beyond the
traditional binary, or radix 2, based system in use today. A greater level of compression is noted in the radix 4 based
system when compared to the radix 2 base as applied to a model of information theory.
A radix 2 based system is composed of two separate character types that have no meaning except
not representing the other character type as defmed by Shannon in 1948. The radix 5 based system
employs five separate characters that have no semantic meaning except not representing the other
characters. Traditional literature has a random string of binary sequential characters as being "less
patterned" than non-random sequential strings. A non-random string of characters will be able to
compress, were as a random string of characters will not be able to compress. This study has
found that a radix 5 based character length allows for equal compression of random and non-
random sequential strings. This has important aspects to information transmission and storage.
The implementation of a ternary or quaternary based system to information infrastructure to replace the
archaic binary system. Using a temary or a quaternary based system will add greater robustness,
compression, and utilizability to future information systems.