Image processing applications typically parallelize well. This gives a developer interested in data throughput
several different implementation options, including multiprocessor machines, general purpose computation on
the graphics processor, and custom gate-array designs. Herein, we will investigate these first two options for
dictionary learning and sparse reconstruction, specifically focusing on the K-SVD algorithm for dictionary learning
and the Batch Orthogonal Matching Pursuit for sparse reconstruction. These methods have been shown to
provide state of the art results for image denoising, classification, and object recognition. We'll explore the GPU
implementation and show that GPUs are not significantly better or worse than CPUs for this application.
Synthetic Aperture Radar (SAR) provides day/night all weather imagery, and as such is being increasingly
utilized for overhead reconnaissance. Additionally, the active, coherent nature of the system provides for analysis
not readily achievable with electro-optical imagery. However, like all coherent systems, SAR imagery suffers
degradation from speckle (a random interference pattern) which hinders interpretation. Herein, we investigate
SAR denoising with a new method based on sparse reconstruction over learned dictionaries and show this
approach performs better than the current state of the art speckle filters.
A single hyperspectral image can easily be hundreds of megabytes or even several gigabytes in size. For spectral
processing, this is not an issue, as each pixel is processed indepedently. Additionally, many standard image
processing algorithms can be readily adapted to process a few lines at a time (possibly with multiple passes
over a file). Thus, clustering algorithms like k-means and dimension reduction methods such as PCA have
also become staples of hyperspectral processing. More recently, however, new algorithms such as locally linear
embedding (LLE) or non-local means have been shown to offer substantial performance increases, at least in
theory. However, incremental processing is not feasible in many of these newer algorithms, and large amounts
of processing power and memory are required to process large images. For this reason, mechanisms to efficiently
reduce the size of large hyperspectral images while maintaining important information are desired. In this paper,
we investigate a segmentation algorithm as a means to this end. We show that the amount of data that needs
to be processed can be reduced by over an order of magnitude while maintaining the spectral purity of the data.
This will be illustrated by showing classification accuracy before and after segmentation.
Ferroelectric and magnetic transducers are utilized in large number of applications, including nanopositioning,
fluid pumps, high-speed milling, and vibration control/suppression. However, the physical mechanisms which
make these materials highly effective transducers inherently introduce nonlinear, hysteretic behavior that must
be incorporated in models and control designs. This significantly complicates control designs and limits the
effectiveness of linear control algorithms when directly applied to the system. One solution is to employ an exact
or approximate inverse model which converts a desired output to the corresponding input. This alleviates the
complex input-output relation, allowing a linear control to be applied. Linearization of the actuator dynamics
in this manner permits subsequent use of linear control designs to achieve high accuracy, high speed tracking as
well as vibration attenuation and positioning objectives.
Reptation, or accommodation, is manifested in ferromagnetic materials in a variety of operating regimes and
hence must be incorporated in models used for comprehensive material characterization or model-based control
design. Because the microscopic mechanisms which cause reptation are complex, we characterize the effect in a
phenomenological macroscopic manner within the context of a homogenized energy framework for ferromagnetic
hysteresis. Attributes of the model are illustrated through comparison with experimental data.
In this paper, we present a new algorithm to implement the homogenized energy hysteresis model with thermal relaxation for both ferroelectric and ferromagnetic materials. The approach conserves most of the accuracy of the original algorithm, but enables all erfc and exp functions to be calculated in advance, thereby requiring that only basic mathematical operations be performed in real time. This is done without a signicant increase in memory usage. Using this approach, execution time of the model has been seen to improve by a factor of 70 for some applications, whereas the error only increases by five ten thousandths (0.05%) of the saturation polarization/magnetization. The model with negligible relaxation is also given, as it is used to illustrate some optimizations. Emphasis is placed on the ecient computation of these models, and theoretical development is left to the references.