The usefulness of orthogonal subspace projection (OSP) has been demonstrated in many applications. Automatic Target
Generation Process (ATGP) was previously developed for automatic target recognition for Hyperspectral imagery by
implementing a successive OSP. However, ATGP itself does not provide a stopping rule to determine how many signal
sources present and need to be extracted in the image. This paper presents a new application of ATGP in determining the
number of signal sources and finding these signal sources in the image at the same time. The idea is to categorize signal
sources into target classes and background classes in terms of their inter-sample spectral correlation (ISSC). Two
separate algorithms, unsupervised target sample generation (UTSG) and unsupervised background sample generation
(UBSG) are developed for this purpose. The UTSG implements a sequence of successive OSP in the sphered
hyperspectral data to determine the number of target signal sources whose ISSC are characterized by high order statistics
(HOS) and find the target signal sources to at the same time. It is then followed by the UBSG which operates the ATGP
on a space orthogonal to the subspace generated by the target samples to determine and find background signal sources.
Both UTSG and UBSG are terminated by an effective stopping rule which can be used to estimate the virtual
dimensionality (VD). Two data sets, synthetic image data and real image scenes are used for experiments. Experimental
results demonstrate that the UTSG and UBSG are effective in extracting signal sources in various applications.
KEYWORDS: Target detection, Hyperspectral imaging, Detection and tracking algorithms, Sensors, Image processing, Data processing, Signal processing, Algorithm development, Signal detection, Data analysis
One of the most challenging issues in unsupervised target analysis is how to obtain unknown target knowledge directly
from the data to be processed. This issue has never arisen in supervised target analysis where the target knowledge is
either assumed to be known or provided by a priori. However, with recent advent of sensor technology many unknown
and subtle signal sources can be uncovered and revealed by high spectral imaging spectrometers such as hyperspectral
imaging sensors. The knowledge of these signal sources generally cannot be obtained by assumed or prior knowledge.
Under this circumstance supervised target analysis may not be realistic or applicable. This paper addresses the issue of
how to generate such knowledge for data analysis and further develops unsupervised target finding algorithms for target
analysis. In order to demonstrate the utility of the developed unsupervised target finding algorithms, experiments are
conducted for applications in unsupervised linear spectral unmixing.
Kernel-based approaches have recently drawn considerable interests in hyperspectral image analysis due to its ability in
expanding features to a higher dimensional space via a nonlinear mapping function. Many well-known detection and
classification techniques such as Orthogonal Subspace Projection (OSP), RX algorithm, linear discriminant analysis,
Principal Components Analysis (PCA), Independent Component Analysis (ICA), have been extended to the
corresponding kernel versions. Interestingly, a target detection method, called Constrained Energy Minimization (CEM)
which has been also widely used in hyperspectral target detection has not been extended to its kernel version. This paper
investigates a kernel-based CEM, called Kernel CEM (K-CEM) which employs various kernels to expand the original
data space to a higher dimensional feature space that CEM can be operated on. Experiments are conducted to perform a
comparative analysis and study between CEM and K-CEM. The results do not show K-CEM provided significant
improvement over CEM in detecting hyperspectral targets but does show significant improvement in detecting targets in
multispectral imagery which provides limited spectral information for the CEM to work well.
This paper presents a wavelet based hierarchical coding scheme for radar image compression. Radar signal is firstly
quantized to digital signal, and reorganized as raster-scanned image according to radar's repeated period frequency. After
reorganization, the reformed image is decomposed to image blocks with different frequency band by 2-D wavelet
transformation, each block is quantized and coded by the Huffman coding scheme. A demonstrating system is
developed, showing that under the requirement of real time processing, the compression ratio can be very high, while
with no significant loss of target signal in restored radar image.
Two major issues encountered in unsupervised hyperspectral image classification are (1) how to determine the number
of spectral classes in the image and (2) how to find training samples that well represent each of spectral classes without
prior knowledge. A recently developed concept, Virtual dimensionality (VD) is used to estimate the number of spectral
classes of interest in the image data. This paper proposes an effective algorithm to generate an appropriate training set
via a recently developed Prioritized Independent Component Analysis (PICA). Two sets of hyperspectral data,
Airborne Visible Infrared Imaging Spectrometer (AVIRIS) Cuprite data and HYperspectral Digital Image Collection
Experiment (HYDICE) data are used for experiments and performance analysis for the proposed method.
This paper presents a texture segmentation approach based on Gauss-Markov random field(GMRF) model and multi-directional mosaics. Image texture is modeled by the second order GMRF model and the least error estimation is employed for the solution of model parameters. In order to improve the segmentation accuracy of uncertain area in boundary region between different textures, we introduced Laws energy masks and directional mosaics to obtain energy and orientation feature. And Euclidean distance approach is employed to classify different features. Experiments show that accuracy of texture segmentation can be improved.