Search

Automated analysis of histological images helps diagnose and further classify breast cancer. Totally automated approaches can be used to pinpoint images for further analysis by the medical doctor. But tissue images are especially challenging for either manual or automated approaches, due to mixed patterns and textures, where malignant regions are sometimes difficult to detect unless they are in very advanced stages. Some of the major challenges are related to irregular and very diffuse patterns, as well as difficulty to define winning features and classifier models. Although it is also hard to segment correctly into regions, due to the diffuse nature, it is still crucial to take low-level features over individualized regions instead of the whole image, and to select those with the best outcomes. In this paper we report on our experiments building a region classifier with a simple subspace division and a feature selection model that improves results over image-wide and/or limited feature sets. Experimental results show modest accuracy for a set of classifiers applied over the whole image, while the conjunction of image division, per-region low-level extraction of features and selection of features, together with the use of a neural network classifier achieved the best levels of accuracy for the dataset and settings we used in the experiments. Future work involves deep learning techniques, adding structures semantics and embedding the approach as a tumor finding helper in a practical Medical Imaging Application.

In this paper, we provide performance analysis of a sparsity-based interpolation technique for direction-of-arrival (DOA) estimation with partially augmentable non-uniform arrays. The degrees-of-freedom (DOFs) offered by a partially augmentable non-uniform array cannot be fully utilized for subspace-based DOA estimation due to the presence of holes in the corresponding difference coarray. The interpolation technique fills the ‘holes’ in the difference coarray, thereby permitting full use of the available DOFs for DOA estimation. We examine the performance of the interpolation-based DOA estimation scheme for scenes with varying source powers under different source separations. Co-prime arrays, which are a type of partially augmentable non-uniform configuration, are utilized for the performance analysis.

Fitting affine objects to data is the basis of many tools and methodologies in statistics, machine learning, and signal processing. The L1 norm is often employed to produce subspaces exhibiting a robustness to outliers and faulty observations. The L1-norm best-fit subspace problem is directly formulated as a nonlinear, nonconvex, and nondifferentiable optimization problem. The case when the subspace is a hyperplane can be solved to global optimality efficiently by solving a series of linear programs. The problem of finding the best-fit line has recently been shown to be NP-hard. We present necessary conditions for optimality for the best-fit subspace problem, and use them to characterize properties of optimal solutions.

Structural health monitoring (SHM) of wind turbines has been applied in the wind energy industry to obtain their real-time vibration parameters and to ensure their optimum performance. For SHM, the accuracy of its results and the efficiency of its measurement methodology and data processing algorithm are the two major concerns. Selection of proper measurement parameters could improve such accuracy and efficiency. The Stochastic Subspace Identification (SSI) is a widely used data processing algorithm for SHM. This research discussed the accuracy and efficiency of SHM using SSI method to identify vibration parameters of on-line wind turbine towers. Proper measurement parameters, such as optimum measurement duration, are recommended.

In this study, the recursive update algorithm of subspace system identification is proposed for estimating dynamic characteristics of a time-varying multi-input multi-output (MIMO) building structure using the pastinput/ output multivariable output-error state space (PO-MOESP) based algorithm. First, the recursive subspace identification with Bona-fide LQ renewing algorithm (RSI-BonaFide-Oblique) incorporated with moving window technique is utilized to identify modal parameters such as natural frequencies, damping ratios and mode shapes at each time instant during the strong earthquake excitation, which assumes the equivalent linear dynamic characteristic in every moving time window with a fixed length. In addition to identifying the instantaneous dynamic characteristics of the structural system using RSI, several damage detection algorithms related to the reconstructed system stiffness matrix using simplified model was derived from the previously identified modal parameters to quantify the damage extent, specify its corresponding damage location, the time of occurrence during the excitation, and the percentage of stiffness reduction. To demonstrate the capability of the proposed recursive subspace identification algorithm and the damage assessment technique, dynamic response data generated from a shaking table test of a two 3-story steel structure experiment is used to verify the algorithms on detecting damage caused by nonlinear behavior occurring on the steel column. Besides, a building seismic response data (collected from Chi-Chi earthquake) was also used to evaluate the damage situation during severe earthquake excitation.

In recent years, with the rapid development of wireless communication industry, how to manage the interference become a central problem in modern communication system. Actually, this problem mostly comes from the less and less spectrum resource and increasing demand for high data rates. Although many interference management techniques have been proposed, because the interference can exist in any part of wireless communication, some basic problems of network interference are not able to be solved until the emergence of interference alignment technology. Interference alignment, in theory, can enable a performance that all the interference signals fall into the subspace of interference and all the useful signal also fall into the corresponding subspace. In this paper, we focus on the problem of signal transmitted over an interference channel, along the lines of the recently proposed methods of interference alignment. From the basic principle of interference alignment, we can see each receiver maintains its corresponding subspace, the transmitters mold their transmissions regularly in order to make all the interference signal received by a particular receiver and then falls into its interference subspace. The remaining part of the receiver space can be used to get the useful signal. For the general interference channel, compared with the previous method, this kind of technique not only minimizes the interference power that is overflowed out the interference subspace, but also minimizes the power of useful signal that is fell into the interference subspace.

Spaceborne sensors systems are characterized by scarce onboard computing and storage resources and by communication links with reduced bandwidth. Random projections techniques have been demonstrated as an effective and very light way to reduce the number of measurements in hyperspectral data, thus, the data to be transmitted to the Earth station is reduced. However, the reconstruction of the original data from the random projections may be computationally expensive. SpeCA is a blind hyperspectral reconstruction technique that exploits the fact that hyperspectral vectors often belong to a low dimensional subspace. SpeCA has shown promising results in the task of recovering hyperspectral data from a reduced number of random measurements. In this manuscript we focus on the implementation of the SpeCA algorithm for graphics processing units (GPU) using the compute unified device architecture (CUDA). Experimental results conducted using synthetic and real hyperspectral datasets on the GPU architecture by NVIDIA: GeForce GTX 980, reveal that the use of GPUs can provide real-time reconstruction. The achieved speedup is up to 22 times when compared with the processing time of SpeCA running on one core of the Intel i7-4790K CPU (3.4GHz), with 32 Gbyte memory.

Hyperspectral remote sensing offers a powerful tool in many different application contexts. The imbalance between the high dimensionality of the data and the limited availability of training samples calls for the need to perform dimensionality reduction in practice. Among traditional dimensionality reduction techniques, feature extraction is one of the most widely used approaches due to its flexibility to transform the original spectral information into a subspace. In turn, band selection is important when the application requires preserving the original spectral information (especially the physically meaningful information) for the interpretation of the hyperspectral scene. In the case of hyperspectral image classification, both techniques need to discard most of the original features/bands in order to perform the classification using a feature set with much lower dimensionality. However, the discriminative information that allows a classifier to provide good performance is usually classdependent and the relevant information may live in weak features/bands that are usually discarded or lost through subspace transformation or band selection. As a result, in practice, it is challenging to use either feature extraction or band selection for classification purposes. Relevant lines of attack to address this problem have focused on multiple feature selection aiming at a suitable fusion of diverse features in order to provide relevant information to the classifier. In this paper, we present a new dimensionality reduction technique, called multiple criteria-based spectral partitioning, which is embedded in an ensemble learning framework to perform advanced hyperspectral image classification. Driven by the use of a multiple band priority criteria that is derived from classic band selection techniques, we obtain multiple spectral partitions from the original hyperspectral data that correspond to several band subgroups with much lower spectral dimensionality as compared with the original band set. An ensemble learning technique is then used to fuse the information from multiple features, taking advantage of the relevant information provided by each classifier. Our experimental results with two real hyperspectral images, collected by the reflective optics system imaging spectrometer (ROSIS) over the University of Pavia in Italy and the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) over the Salinas scene, reveal that our presented method, driven by multiple band priority criteria, is able to obtain better classification results compared with classic band selection techniques. This paper also discusses several possibilities for computationally efficient implementation of the proposed technique using various high-performance computing architectures.

Digital holography has received recent attention for many imaging and sensing applications, including imaging through turbulent and turbid media, adaptive optics, three-dimensional projective display technology and optical tweezing. It holds several advantages over conventional imaging and wavefront sensing, chief among these being significantly fewer and simpler optical components and the retrieval of complex field. A significant obstacle for digital holography in real-time applications, such as wavefront sensing for high-energy laser systems and high-speed imaging for target tracking, is the fact that digital holography is computationally intensive; it requires iterative virtual wavefront propagation and hill-climbing algorithms to optimize sharpness criteria. This research demonstrates real-time methods for digital holography based on approaches for optimal and adaptive identification, prediction, and control of optical wavefronts. The methods presented integrate minimum-variance wavefront prediction into dynamic digital holography schemes to accelerate the wavefront correction and image sharpening algorithms. Further gains in computational efficiency are demonstrated in this work with a variant of localized sharpening in conjunction with predictive dynamic digital holography for real-time applications. This "subspace correction" method optimizes sharpness of local regions in a detector plane by parallel independent wavefront correction on reduced-dimension subspaces of the complex field in a spectral plane.

Latent low-rank representation (LatLRR) has been attached considerable attention in the field of remote sensing image segmentation, due to its effectiveness in exploring the multiple subspace structures of data. However, the increasingly heterogeneous texture information in the high spatial resolution remote sensing images, leads to more severe interference of pixels in local neighborhood, and the LatLRR fails to capture the local complex structure information. Therefore, we present a local sparse structure constrainted latent low-rank representation (LSSLatLRR) segmentation method, which explicitly imposes the local sparse structure constraint on LatLRR to capture the intrinsic local structure in manifold structure feature subspaces. The whole segmentation framework can be viewed as two stages in cascade. In the first stage, we use the local histogram transform to extract the texture local histogram features (LHOG) at each pixel, which can efficiently capture the complex and micro-texture pattern. In the second stage, a local sparse structure (LSS) formulation is established on LHOG, which aims to preserve the local intrinsic structure and enhance the relationship between pixels having similar local characteristics. Meanwhile, by integrating the LSS and the LatLRR, we can efficiently capture the local sparse and low-rank structure in the mixture of feature subspace, and we adopt the subspace segmentation method to improve the segmentation accuracy. Experimental results on the remote sensing images with different spatial resolution show that, compared with three state-of-the-art image segmentation methods, the proposed method achieves more accurate segmentation results.

The remotely sensed reflectance spectra of vegetated surfaces contain information relating to the leaf area index (LAI) and the chlorophyll-a and -b concentrations (Cab) in a leaf. Difficulties associated with the retrieval of these two biophysical parameters from a single reflectance spectrum arise mainly from the choice of a suitable set of observation wavelengths and the development of a retrieval algorithm. Efforts have been applied toward the development of new algorithms, such as the numerical inversion of radiative transfer models, in addition to the development of simple approaches based on the spectral vegetation indices. This study explored a different approach: An equation describing band-to-band relationships (vegetation isoline equation) was used to retrieve the LAI and C ab simultaneously from a reflectance spectrum. The algorithm used three bands, including the red edge region, and an optimization cost function was constructed from two vegetation isoline equations in the red-NIR and red edge-NIR reflectance subspaces. A series of numerical experiments was conducted using the PROSPECT model to explore the numerical challenges associated with the use of the vegetation isoline equation during the parameter retrieval of the LAI and C ab . Overall, our results indicated the existence of a global minimum (and no local minima) over a wide swath of the LAI-C ab parameter subspace in most simulation cases. These results suggested that the use of the vegetation isoline equation in the simultaneous retrieval of the LAI and the C ab provides a viable alternative to the spectral vegetation index algorithms and the direct inversion of the canopy radiative transfer models.

Information on the relationships between pairs of wavelength bands is useful when analyzing multispectral sensor data. The soil isoline is one such relationship that is obtained under a constant soil spectrum. However, numerical determination of the soil isoline in the red and NIR reflectance subspace is problematic because of singularities encountered during polynomial fitting. In our previous work, this difficulty was effectively overcome by rotating the original red-NIR subspace by an angle identical to a soil line slope. In the context of hyperspectral data analysis, the applicability of this approach should be investigated thoroughly for band combinations other than red-NIR. The objective of the present study was to expand the applicable range of band combinations to 400–2500 nm by conducting a set of numerical simulations of radiative transfer. Soil isolines were determined numerically by varying soil reflectance and biophysical parameters. The results demonstrated that, as shown previously for the red-NIR band combination, singularities can be avoided for most band combinations through use of the rotation approach. However, for some combinations, especially those involving the shortwave infrared range, the rotation approach gave rise to a further numerical singularity. The present findings thus indicate that special caution should be exercised in the numerical determination of soil isoline equations when one of the chosen bands is in the shortwave infrared region.

Nonuniformity in FPAs (focal plane arrays) is a common, although undesirable, characteristic arising from small differences in the responsivity of individual detectors. Now there are many scene-based nonuniformity correction algorithms. In these correction algorithms, interframe registration based nonuniformity correction algorithm is a principal type. The core of interframe registration based nonuniformity correction algorithm is accurate image registration. However, because of the existence of the nonuniformity, the values of the adjacent columns or rows of the image with noise are so close that it is difficult to distinguish them. We find it is difficult to achieve the corresponding rows or columns registration between two frames accurately. It will lead to the wrong registration that will affect the nonuniformity correction effect. In this method, we propose a nonuniformity correction algorithm based on subspace projection registration. By projecting the rows and columns of the image on to their corresponding subspace respectively, we can realize the dimension reduction of the row space and the column space. As a result, the value differences between the adjacent rows or the adjacent columns of the image can be amplified. Then the algorithm can identify the corresponding rows and columns between two frames, enabling accurate image registration. Experiments demonstrate that our algorithm achieve accurate image registration and outstanding correction effect.

Conventionally, SAR Automatic Target Detection(ATD)are often performed in SAR image domain. A novel scheme of SAR target detection in the state of non-imaging is presented. More precisely, this paper addresses adaptive detection of SAR possibly extended targets when implemented on range-compressed but azimuth-uncompressed SAR raw data. The SAR target detection is established in the context of space-time adaptive processing (STAP) and the spatial-temporal steering vector of an airborne stripmap SAR is derived by exploiting signature diversity, namely of the fact that SAR can change the transmitted signal as the azimuth varies. The generalized adaptive subspace detector (GASD) is employed to detect range-spread target from SAR raw data. Performance analysis of the proposed detector via Monte Carlo simulation shows the validity of this new detection scheme.

Traditional subspace detection for the second-order Gaussian (SOG) model signal is generally considered in the homogeneous or partially homogeneous environments. This paper addresses the problem of the subspace detection for the SOG signal in the presence of the nonhomogeneous noise whose covariance matrices in the primary and secondary data are assumed to be random, with some appropriate distributions. Within this nonhomogeneous framework, a novel adaptive subspace detector is proposed in terms of an approximate generalized likelihood ratio test (AGLRT) and the Gibbs sampling strategy. The numerical result evaluates the performance of the subspace detector with Monte Carlo method under nonhomogeneity.

When medical data are collected there are many Vital Sign Signals (VSSs) that can be used for data analysis. From a hyperspectral imaging perspective, we can consider a patient with different vital sign signals as a pixel vector in hyperspectral image and each vital sign signal as a particular band. In light of this interpretation this paper develops two new concepts of prioritization of VSSs. One is Orthogonal Subspace Projection Residual (OSPR), which measures the residual of a VSS in the orthogonal complement subspace to the space linearly spanned by the remaining VSSs. Another is to construct a histogram for each of VSSs that can be used as a means of ranking VSSs according to a certain criterion for optimality. Several measures are proposed to be used as criteria for VSS prioritization, which are variance, entropy and Kullbak-Leibler (KL) information measure. VSS prioritization can then be used as the VSS selection method to form Logistic Regression model (LRM). In order to determine how many VSSs should be used a recently developed concept, called Virtual Dimensionality (VD) can be used for this purpose. To demonstrate the utility of VSS prioritization, data collected in University of Maryland, School of Medicine, Shock Trauma Center (STC) was used for experiments.

In discriminating target materials from background clutter in hyperspectral imagery, one must contend with variability in both. Most algorithms focus on the clutter variability, but for some materials there is considerable variability in the spectral signatures of the target. This is especially the case for solid target materials, whose signatures depend on morphological properties (particle size, packing density, etc.) that are rarely known a priori. In this paper, we investigate detection algorithms that explicitly take into account the diversity of signatures for a given target. In particular, we investigate variable target detectors when applied to new representations of the hyperspectral data: a manifold learning based approach, and a residual based approach. The graph theory and manifold learning based approach incorporates multiple spectral signatures of the target material of interest; this is built upon previous work that used a single target spectrum. In this approach, we first build an adaptive nearest neighbors (ANN) graph on the data and target spectra, and use a biased locally linear embedding (LLE) transformation to perform nonlinear dimensionality reduction. This biased transformation results in a lower-dimensional representation of the data that better separates the targets from the background. The residual approach uses an annulus based computation to represent each pixel after an estimate of the local background is removed, which suppresses local backgrounds and emphasizes the target-containing pixels. We will show detection results in the original spectral space, the dimensionality-reduced space, and the residual space, all using subspace detectors: ranked spectral angle mapper (rSAM), subspace adaptive matched filter (ssAMF), and subspace adaptive cosine/coherence estimator (ssACE). Results of this exploratory study will be shown on a ground-truthed hyperspectral image with variable target spectra and both full and mixed pixel targets.

Remotely sensed data fusion aims to integrate multi-source information generated from different perspectives, acquired with different sensors or captured at different times in order to produce fused data that contains more information than one individual data source. Recently, extended morphological attribute profiles (EMAPs) were proposed to embed contextual information, such as texture, shape, size and etc., into a high dimensional feature space as an alternative data source to hyperspectral image (HSI). Although EMAPs provide greater capabilities in modeling both spatial and spectral information, they lead to an increase in the dimensionality of the extracted features. Conventionally, a data point in high dimensional feature space is represented by a vector. For HSI, this data representation has one obvious shortcoming in that only spectral knowledge is utilized without contextual relationship being exploited. Tensors provide a natural representation for HSI data by incorporating both spatial neighborhood awareness and spectral information. Besides, tensors can be conveniently incorporated into a superpixel-based HSI image processing framework. In our paper, three tensor-based dimensionality reduction (DR) approaches were generalized for high dimensional image with promising results reported. Among the tensor-based DR approaches, the Tensor Locality Preserving Projection (TLPP) algorithm utilized graph Laplacian to model the pairwise relationship among the tensor data points. It also demonstrated excellent performance for both pixel-wise and superpixel-wise classification on Pavia University dataset.

In this work, array processing techniques based on subspace decomposition of signal have been evaluated for estimation of direction of arrival of moving targets using acoustic signatures. Three subspace based approaches – Incoherent Wideband Multiple Signal Classification (IWM), Least Square-Estimation of Signal Parameters via Rotation Invariance Techniques (LS-ESPRIT) and Total Least Square- ESPIRIT (TLS-ESPRIT) are considered. Their performance is compared with conventional time delay estimation (TDE) approaches such as Generalized Cross Correlation (GCC) and Average Square Difference Function (ASDF). Performance evaluation has been conducted on experimentally generated data consisting of acoustic signatures of four different types of civilian vehicles moving in defined geometrical trajectories. Mean absolute error and standard deviation of the DOA estimates w.r.t. ground truth are used as performance evaluation metrics. Lower statistical values of mean error confirm the superiority of subspace based approaches over TDE based techniques. Amongst the compared methods, LS-ESPRIT indicated better performance.

In many practical situations visual pattern recognition is vastly burdened by low quality of input images due to noise, geometrical distortions, as well as low quality of the acquisition hardware. However, although there are techniques of image quality improvements, such as nonlinear filtering, there are only few attempts reported in the literature that try to build these enhancement methods into a complete chain for multi-dimensional object recognition such as color video or hyperspectral images. In this work we propose a joint multilinear signal filtering and classification system built upon the multi-dimensional (tensor) approach. Tensor filtering is performed by the multi-dimensional input signal projection into the tensor subspace spanned by the best-rank tensor decomposition method. On the other hand, object classification is done by construction of the tensor sub-space constructed based on the Higher-Order Singular Value Decomposition method applied to the prototype patters. In the experiments we show that the proposed chain allows high object recognition accuracy in the real-time even from the poor quality prototypes. Even more importantly, the proposed framework allows unified classification of signals of any dimensions, such as color images or video sequences which are exemplars of 3D and 4D tensors, respectively. The paper discussed also some practical issues related to implementation of the key components of the proposed system.

In this paper, we introduce and study a novel unsupervised domain adaptation (DA) algorithm, called latent subspace sparse representation based domain adaptation, based on the fact that source and target data that lie in different but related low-dimension subspaces. The key idea is that each point in a union of subspaces can be constructed by a combination of other points in the dataset. In this method, we propose to project the source and target data onto a common latent generalized subspace which is a union of subspaces of source and target domains and learn the sparse representation in the latent generalized subspace. By employing the minimum reconstruction error and maximum mean discrepancy (MMD) constraints, the structure of source and target domain are preserved and the discrepancy is reduced between the source and target domains and thus reflected in the sparse representation. We then utilize the sparse representation to build a weighted graph which reflect the relationship of points from the different domains (source-source, source- target, and target-target) to predict the labels of the target domain. We also proposed an efficient optimization method for the algorithm. Our method does not need to combine with any classifiers and therefore does not need train the test procedures. Various experiments show that the proposed method perform better than the competitive state of art subspace-based domain adaptation.

Anomaly detection (AD) becomes increasingly important in hyperspectral imagery analysis with many practical applications. Local orthogonal subspace projection (LOSP) detector is a popular anomaly detector which exploits local endmembers/eigenvectors around the pixel under test (PUT) to construct background subspace. However, this subspace only takes advantage of the spectral information, but the spatial correlat ion of the background clutter is neglected, which leads to the anomaly detection result sensitive to the accuracy of the estimated subspace. In this paper, a local three dimensional orthogonal subspace projection (3D-LOSP) algorithm is proposed. Firstly, under the jointly use of both spectral and spatial information, three directional background subspaces are created along the image height direction, the image width direction and the spectral direction, respectively. Then, the three corresponding orthogonal subspaces are calculated. After that, each vector along three direction of the local cube is projected onto the corresponding orthogonal subspace. Finally, a composite score is given through the three direction operators. In 3D-LOSP, the anomalies are redefined as the target not only spectrally different to the background, but also spatially distinct. Thanks to the addition of the spatial information, the robustness of the anomaly detection result has been improved greatly by the proposed 3D-LOSP algorithm. It is noteworthy that the proposed algorithm is an expansion of LOSP and this ideology can inspire many other spectral-based anomaly detection methods. Experiments with real hyperspectral images have proved the stability of the detection result.

In this paper, the noncentral chi-squared distribution is applied in the Constant False Alarm Rate (CFAR) detection of hyperspectral projected images to distinguish the anomaly points from background. Usually, the process of the hyperspectral anomaly detectors can be considered as a linear projection. These operators are linear transforms and their results are quadratic form which comes from the transform of spectral vector. In general, chi-squared distribution could be the proper choice to describe the statistical characteristic of this projected image. However, because of the strong correlation among the bands, the standard central chi-squared distribution often cannot fit the stochastic characteristic of the projected images precisely. In this paper, we use a noncentral chi-squared distribution to approximate the projected image of subspace based anomaly detectors. Firstly, the statistical modal of the projected multivariate data is analysed, and a noncentral chi-squared distribution is deduced. Then, the approach of the parameters calculation is introduced. At last, the aerial hyperspectral images are used to verify the effectiveness of the proposed method in tightly modeling the projected image statistic distribution.

A temporal method to detect infrared small target using recursive sparse recovery is presented. Three advantages are provided by this method: detection of multiscale targets, removing false alarms caused by complex clutter, less computational complexity. Firstly, an initial estimate of the background subspace is trained. Secondly, the target and the background are recovered by low-rank and sparse decomposition based on trained background subspace. Thirdly, the background subspace is updated by background estimation. Finally, the small target can be easily extracted by the threshold segmentation. Experimental results indicate that the proposed method is robust and efficient in complex backgrounds.

Indoor Visible Light Communication (VLC) positioning is an new indoor positioning method. We discuss the method to realize 3D positioning and the how to optimize it. Build a simplified 3D fingerprint database in the condition of linear light intensity attenuation model and use it for 3D positioning. Furthermore we combined fingerprint database method with the iterative algorithm of Lambert distribution to calculate Z coordinates in order to improve the accuracy. Simulations shows that, in the subspace of 1m*1m*1m, the measurement error of fingerprint database method in X-Y plane is 4.1cm, while z direction measurement error is 5.4cm. By contrast, measurement error of 3D positioning method based on iterative algorithm reduces by 10%.

Detection of a subspace anomaly is an important application of hyperspectral imaging in remote sensing. Sub-space anomaly detection depends on the unknown dimension of the main background subspace. When the dimension is high, detection algorithms tend to have unsatisfactory performance. This paper proposes an anomaly detection algorithm that will continue to perform satisfactorily when the dimension is high.

Two-dimensional (2D) transmit beamforming aims at focusing the transmitted energy within certain desired sector while minimizing the amount of energy in the out-of-sector regions. In this paper, we propose parsimonious formulations to the sidelobe control problem in 2D transmit beamforming with multidimensional arrays. The out-of-sector region is partitioned into a small number of subsectors where the subspace spanned by the steering vectors associated with the spatial directions within a certain subsecetor is approximated by the effective discrete- prolate spheroidal sequences associated with that subsector. Then, the sidelobe control is achieved by imposing constraints on the magnitude of the inner product between the 2D transmit beamforming weight vector and the discrete-prolate spheroidal sequences. Simulations examples are presented which show the effectiveness of the proposed formulations.

The objective of this study is to demonstrate the application of two different system identification methods on the structural health monitoring of a bridge. The numerical simulation of bridge-vehicle interaction with road surface roughness is considered in this study for system identification. To identify the bridge dynamic characteristics Covariance-driven Stochastic Subspace Identification method (SSI-COV) in cooperated with Wavelet Packet Transform (WPT) decomposition are used to extract the natural frequencies and mode shapes of the system. For comparison, a popular blind source separation technique called Second Order Blind Identification (SOBI) is also used. Comparison between these two different identification methods is discussed. It was demonstrated that the bridge natural frequencies can be identified by the proposed two system identification techniques. Besides, the SOBI algorithm can avoid the difficulty of determining of parameters by using SSI-COV algorithm, such as system order, row of Hankel matrix, etc. Finally, a damage scenario of the bridge structure is provided and damage detection algorithms are also proposed to quantify and locate the damage.

Stochastic subspace identification method (SSI) has been proven to be an efficient algorithm for the identification of liner-time-invariant system using multivariate measurements. Generally, the estimated modal parameters through SSI may be afflicted with statistical uncertainty, e.g. undefined measurement noises, non-stationary excitation, finite number of data samples etc. Therefore, the identified results are subjected to variance errors. Accordingly, the concept of the stabilization diagram can help users to identify the correct model, i.e. through removing the spurious modes. Modal parameters are estimated at successive model orders where the physical modes of the system are extracted and separated from the spurious modes. Besides, an uncertainty computation scheme was derived for the calculation of uncertainty bounds for modal parameters at some given model order. The uncertainty bounds of damping ratios are particularly interesting, as the estimation of damping ratios are difficult to obtain. In this paper, an automated stochastic subspace identification algorithm is addressed. First, the identification of modal parameters through covariance-driven stochastic subspace identification from the output-only measurements is used for discussion. A systematic way of investigation on the criteria for the stabilization diagram is presented. Secondly, an automated algorithm of post-processing on stabilization diagram is demonstrated. Finally, the computation of uncertainty bounds for each mode with all model order in the stabilization diagram is utilized to determine system natural frequencies and damping ratios. Demonstration of this study on the system identification of a three-span steel bridge under operation condition is presented. It is shown that the proposed new operation procedure for the automated covariance-driven stochastic subspace identification can enhance the robustness and reliability in structural health monitoring.

In this paper, we propose a novel palmprint recognition algorithm. Firstly, the palmprint images are represented by the anisotropic filter. The filters are built on Gaussian functions along one direction, and on second derivative of Gaussian functions in the orthogonal direction. Also, this choice is motivated by the optimal joint spatial and frequency localization of the Gaussian kernel. Therefore,they can better approximate the edge or line of palmprint images. A palmprint image is processed with a bank of anisotropic filters at different scales and rotations for robust palmprint features extraction. Once these features are extracted, subspace analysis is then applied to the feature vectors for dimension reduction as well as class separability. Experimental results on a public palmprint database show that the accuracy could be improved by the proposed novel representations, compared with Gabor.

In this paper we present a method for online 3D face reconstruction from a video sequence. The face landmarks in a given frame are detected and used to create a 3D shape estimate. The resulting 3D shape is an approximate, sparse representation of the subject’s face. Our reconstruction step is based on a revised version of incremental Structure From Motion, where we use a novel 4D subspace tracking procedure followed by scaled deflation against a vector of ones. Facial landmark detection is built upon a regressor cascade scheme where each subsequent regressor updates the initial shape obtained from the preceding frame.

Shape matching and recognition is a challenging task due to geometric distortions and occlusions. A novel shape matching approach based on Grassmann manifold is proposed that affine transformations and partial occlusions are both considered. An affine invariant Grassmann shape descriptor is employed which projects one shape contour to a point on Grassmann manifold and gives the similarity measurement between two contours based on the geodesic distance on the manifold. At first, shape contours are parameterized by affine length and accordingly divided into local affine-invariant shape segments, which are represented by the Grassmann shape descriptor, according to their curvature scale space images. Then the Smith-Waterman algorithm is employed to find the common parts of two shapesr segment sequences, and get the local similarity of shapes. The global similarity is given by the found common parts, and finally the shape recognition accomplished by the weighted sum of local similarity and global similarity. The robustness of the Grassmann shape descriptor is analyzed through subspace perturbation analysis theory. Retrieval experiments show that our approach is effective and robust under affine transformations and partial occlusions.

The orithogonal subspace projection (OSP) method needs all the endmember spectral information of observation area which is usually unavailable in actual situation. In order to extend the application of OSP method, this paper proposes an algorithm without any priori information based on OSP. Firstly, the background endmember spectral matrix is obtained by using unsupervised method. Then, the OSP projection operator is calculated with the background endmember matrix. Finally, the detection operator is constructed by using the projection operator, which is used to detect the hyperspectral imagery pixel by pixel. In order to increase the detection rate, local processing is proposed for anomaly detection with no prior knowledge. The algorithm is tested with AVIRIS hyperspectral data, and binary image of targets and ROC curves are given in the paper. Experimental results show that the proposed anomaly detection method based on OSP is more effective than the classic RX detection algorithm under the case of insufficient prior knowledge, and the detection rate is significantly increased by using the local processing.

In this paper, we propose a robust tracking method for infrared object. We introduce the appearance model and the sparse representation in the framework of particle filter to achieve this goal. Representing every candidate image patch as a linear combination of bases in the subspace which is spanned by the target templates is the mechanism behind this method. The natural property, that if the candidate image patch is the target so the coefficient vector must be sparse, can ensure our algorithm successfully. Firstly, the target must be indicated manually in the first frame of the video, then construct the dictionary using the appearance model of the target templates. Secondly, the candidate image patches are selected in following frames and the sparse coefficient vectors of them are calculated via `1-norm minimization algorithm. According to the sparse coefficient vectors the right candidates is determined as the target. Finally, the target templates update dynamically to cope with appearance change in the tracking process. This paper also addresses the problem of scale changing and the rotation of the target occurring in tracking. Theoretic analysis and experimental results show that the proposed algorithm is elective and robust.

Hyper-spectral remote sensing data can be acquired by imaging the same area with multiple wavelengths, and it normally consists of hundreds of band-images. Hyper-spectral images can not only provide spatial information but also high resolution spectral information, and it has been widely used in environment monitoring, mineral investigation and military reconnaissance. However, because of the corresponding large data volume, it is very difficult to transmit and store Hyper-spectral images. Hyper-spectral image dimensional reduction technique is desired to resolve this problem. Because of the High relation and high redundancy of the hyper-spectral bands, it is very feasible that applying the dimensional reduction method to compress the data volume. This paper proposed a novel band selection-based dimension reduction method which can adaptively select the bands which contain more information and details. The proposed method is based on the principal component analysis (PCA), and then computes the index corresponding to every band. The indexes obtained are then ranked in order of magnitude from large to small. Based on the threshold, system can adaptively and reasonably select the bands. The proposed method can overcome the shortcomings induced by transform-based dimension reduction method and prevent the original spectral information from being lost. The performance of the proposed method has been validated by implementing several experiments. The experimental results show that the proposed algorithm can reduce the dimensions of hyper-spectral image with little information loss by adaptively selecting the band images.

In this paper, modern subspace identification methods are applied to a Multi-conjugate Adaptive optics Demonstrator, MAD, developed at ESO. The identified multi-input multi-output systems mapping voltages into slopes can be obtained on data taken in open/closed loop on beacon and in both SCAO and GLAO configurations. The advantages of the proposed approach is twofold: on the one hand the experiment to collect the data takes only few minutes during day time, it can be done on beacon, and all the computational effort is moved off-line. On the other hand, subspace identification provides the mathematical model necessary to design modern model-based controllers (e.g. linear quadratic control).

Conventional subspace-based signal direction-of-arrival estimation methods rely on the familiar L 2 -norm-derivedprincipal components (singular vectors) of the observed sensor-array data matrix. In this paper, for the “rsttime in the literature, we “nd the L 1 -norm maximum projection components of the observed data and searchin their subspace for signal presence. We demonstrate that L 1 -subspace direction-of-arrival estimation exhibits(i) similar performance to L 2 (usual singular-value/eigen-vector decomposition) direction-of-arrival estimationunder normal nominal-data system operation and (ii) signi“cant resistance to sporadic/occasional directionaljamming and/or faulty measurements.Keywords: Dimensionality reduction, direction-of-arrival estimation, erroneous data, faulty measurements,jamming, L 1 norm, L 2 norm, principal-component analysis, outlier resistance, subspace signal processing. 1. INTRODUCTION Direction-of-arrival (DoA) estimation is a fundamental problem in signal processing with important applicationsin radar, sonar, source localization, wireless communications, and much more.

Search algorithm complexity modeling, in the form of penetration rate estimation, provides a useful way to estimate search efficiency in application domains which involve searching over a hypothesis space of reference templates or models, as in model-based object recognition, automatic target recognition, and biometric recognition. The penetration rate quantifies the expected portion of the database that must be searched, and is useful for estimating search algorithm computational requirements. In this paper we perform mathematical modeling to derive general equations for penetration rate estimates that are applicable to a wide range of recognition problems. We extend previous penetration rate analyses to use more general probabilistic modeling assumptions. In particular we provide penetration rate equations within the framework of a model-based image alignment application domain in which a prioritized hierarchical grid search is used to rank subspace bins based on matching probability. We derive general equations, and provide special cases based on simplifying assumptions. We show how previously-derived penetration rate equations are special cases of the general formulation. We apply the analysis to model-based logo image alignment in which a hierarchical grid search is used over a geometric misalignment transform hypothesis space. We present numerical results validating the modeling assumptions and derived formulation.

This paper presents a subspace system identification for estimating the stiffness matrix and flexural rigidities of a shear building under earthquake. Subspace SI is a kind of inverse problem and suffers from inherent instabilities caused by modeling error and measurement noise. The size of Hankel matrix (k(m+p)×T w /Δt), which represents the amount of selected dynamic data among measured responses, is closely related to the accuracy and numerical instability of estimated system matrices. The numerical instability and accuracy of subspace SI is investigated through the estimation error curve of stiffness matrix. The estimation error curve is obtained with respect to the number of block row(k) and sampling rate (Δt) for various time window size (Tw) using a prior finite element model of a shear building. k, Δt and T w resulting in a target accuracy level, are determined through this curve considering the computational cost of subspace identification. The validity of the proposed method is demonstrated through the numerical example of a five-story shear building model with and without damage.

While the plethora of information from multiple imaging and non-imaging data streams presents an opportunity for discovery of fused multimodal, multiscale biomarkers, they also introduce multiple independent sources of noise that hinder their collective utility. The goal of this work is to create fused predictors of disease diagnosis and prognosis by combining multiple data streams, which we hypothesize will provide improved performance as compared to predictors from individual data streams. To achieve this goal, we introduce supervised multiview canonical correlation analysis (sMVCCA), a novel data fusion method that attempts to find a common representation for multiscale, multimodal data where class separation is maximized while noise is minimized. In doing so, sMVCCA assumes that the different sources of information are complementary and thereby act synergistically when combined. Although this method can be applied to any number of modalities and to any disease domain, we demonstrate its utility using three datasets. We fuse (i) 1.5 Tesla (T) magnetic resonance imaging (MRI) features with cerbrospinal fluid (CSF) proteomic measurements for early diagnosis of Alzheimer’s disease (n = 30), (ii) 3T Dynamic Contrast Enhanced (DCE) MRI and T2w MRI for in vivo prediction of prostate cancer grade on a per slice basis (n = 33) and (iii) quantitative histomorphometric features of glands and proteomic measurements from mass spectrometry for prediction of 5 year biochemical recurrence postradical prostatectomy (n = 40). Random Forest classifier applied to the sMVCCA fused subspace, as compared to that of MVCCA, PCA and LDA, yielded the highest classification AUC of 0.82 +/- 0.05, 0.76 +/- 0.01, 0.70 +/- 0.07, respectively for the aforementioned datasets. In addition, sMVCCA fused subspace provided 13.6%, 7.6% and 15.3% increase in AUC as compared with that of the best performing individual view in each of the three datasets, respectively. For the biochemical recurrence dataset, Kaplan-Meier curves generated from classifier prediction in the fused subspace reached the significance threshold (p = 0.05) for distinguishing between patients with and without 5 year biochemical recurrence, unlike those generated from classifier predictions of the individual modalities.

This work investigates the problem of anomaly detection by means of an agnostic inference strategy based on the concepts of spatial saliency and data sparsity. Specifically, it addresses the implementation and experimental validation aspects of a salient feature extraction methodology that was recently proposed for laser-based diagnostics and leverages the wavefield spatial reconstruction capability offered by scanning laser vibrometers. The methodology consists of two steps. The first is a spatiotemporal windowing strategy designed to partition the structural domain in small sub-domains and replicate impinging wave conditions at each location. The second is the construction of a low-rank-plus-outlier model of the regional data set using principal component analysis. Regions are labeled salient when their behavior does not belong to a common low-dimensional subspace that successfully describes the typical behavior of the anomaly-free portion of the surrounding medium. The most at­ tractive feature of this method is that it requires virtually no knowledge of the structural and material properties of the medium. This property makes it a powerful diagnostic tool for the inspection of media with pronounced heterogeneity or with unknown or unreliable material property distributions, e.g., as a result of severe material degradation over large portions of their domain.

Different face views project different face topology in 2D images. The unified processing of face images with less topology different related to smaller range of face view angles is more convenient, and vice versa. Thus many researches divided the entire face pattern space form multiview face images into many subspaces with small range of view angles. However, large number of subspaces is computationally demanding, and different face processing algorithms take different strategies to handle the view changing. Therefore, the research of proper division of face pattern space is needed to ensure good performance. Different from other researches, this paper proposes an optimal view angle range criterion of face pattern space division in theory by careful analysis on the structure differences of multiview faces and on the influence to face processing algorithms. Then a face pattern space division method is proposed. Finally, this paper uses the proposed criterion and method to divide the face pattern space for face detection and compares with other division results. The final results show the proposed criterion and method can satisfy the processing performance with minimum number of subspace. The study in this paper can also help other researches which need to divide pattern space of other objects based on their different views.

The problem of estimating the state of thrusting/ballistic endoatmospheric projectiles moving in 3-dimensional (3-D) space using 2-dimensional (2-D) measurements from a single passive sensor is investigated. The location of projectile’s launch point (LP) is unavailable and this could significantly affect the performance of the estimation and the IPP. The LP altitude is then an unknown target parameter. The estimability is analyzed based on the Fisher Information Matrix (FIM) of the target parameter vector, comprising the initial launch (azimuth and elevation) angles, drag coefficient, thrust and the LP altitude, which determine the trajectory according to a nonlinear motion equation. The full rank of the FIM ensures that one has an estimable target parameters. The corresponding Cram´er-Rao lower bound (CRLB) quantifies the estimation performance of the estimator that is statistically efficient and can be used for IPP. In view of the inherent nonlinearity of the problem, the maximum likelihood (ML) estimate of the target parameter vector is found by using a mixed (partially grid-based) search approach. For a selected grid in the drag-coefficient-thrust-altitude subspace, the proposed parallelizable approach is shown to have reliable estimation performance and further leads to the final IPP of high accuracy.

Dimensionality Reduction (DR) for hyperspectral image data can be regarded as a problem of signal subspace estimation (SSE) in terms of the Linear Mixing Model (LMM). Most SSE methods for hyperspectral data are based on the analysis of second-order statistics (SOS) without considering preservation of anomalies. This paper addresses the problem of SSE for preserving both abundant and rare signal components in hyperspectral images. The multivariate sample skewness for testing normality is brought in our new algorithm as a discrimination index for rank determination of rare vectors subspace, combining with analysis of the maximum of data-residual l 2 -norm denoted as l 2,∞- norm which is strongly influenced by the anomaly signal components. And the SOS based method, labeled as hyperspectral signal subspace identification by minimum error (HySime), is employed for identification of abundant vectors space. The results of experiments on real AVIRIS data prove that multivariate sample skewness statistics is suitable for measuring the distribution about hyperspectral data globally, and our algorithm can obtain the anomaly components from data that are discarded by HySime, which implies less information loss in the our method.

We look at the design of projective measurements for compressive imaging based upon image priors and device constraints. If one assumes that image patches from natural imagery can be modeled as a low rank manifold, we develop an optimality criterion for a measurement matrix based upon separating the canonical elements of the manifold prior. We then describe a stochastic search algorithm for finding the optimal measurements under device constraints based upon a subspace mismatch algorithm. The algorithm is then tested on a prototype compressive imaging device designed to collect an 8x4 array of projective measurements simultaneously. This work is based upon work supported by DARPA and the SPAWAR System Center Pacific under Contract No. N66001-11-C-4092. The views expressed are those of the author and do not reflect the official policy or position of the Department of Defense or the U.S. Government.

The estimation of the covariance matrix from real data is required in the application of space-time adaptive processing (STAP) to an airborne ground moving target indication (GMTI) radar. A natural approach to estimation of the covariance matrix that is based on the information geometry has been proposed. In this paper, the output of the Riemannian mean is used in inversion and projection algorithms. It is found that the projection class of algorithms can yield very significant gains, even when the gains due to inversion-based algorithms are marginal over standard algorithms. The performance of the projection class of algorithms does not appear to be overly sensitive to the projected subspace dimension.

Performance of algorithms for target signal detection in Hyperspectral Imagery (HSI) is often deteriorated when the data is neither statistically homogeneous nor Gaussian or when its Joint Probability Density (JPD) does not match any presumed particular parametric model. In this paper we propose a novel detection algorithm which first attempts at dividing data domain into mostly Gaussian and mostly Non-Gaussian (NG) subspaces, and then estimates the JPD of the NG subspace with a non-parametric Graph-based estimator. It then combines commonly used detection algorithms operating on the mostly-Gaussian sub-space and an LRT calculated directly with the estimated JPD of the NG sub-space, to detect anomalies and known additive-type target signals. The algorithm performance is compared to commonly used algorithms and is found to be superior in some important cases.

Automatic and real-time face recognition can be applied into many attractive applications. For example, at a checkpoint it is expected that there are no burdens on a passing person and a security guard in addition to low cost. Normally a unique 3D person is projected into 2D images with information loss. It means a person is no longer unique in 2D space. Furthermore the various conditions such as pose variance, illumination variance and different expression make face recognition difficult. In order to separate a person, his or her subspace should have several faces and be redundant. That is why the database naturally becomes large. Under this situation the efficient face recognition is a key to a surveillance system. Face recognition by spars representation classification (SRC) could be one of promising candidates to realize rapid face recognition. This method can be understood in a similar way to compressive sensing (CS). In this paper, we propose the efficient approach of face recognition by SRC for multiple poses from the viewpoint of CS. The part-cropped database (PCD) is suggested to avoid position misalignments by discarding the information of topological linkages among eyes, a nose and a mouth. Although topological linkages are important for face recognition in general, they cause position misalignments among multiple poses which decrease recognition rate. Our approach solves one of trade-off problem between keeping topological linkages and avoiding position misalignments. According to the simulated experiments, PCD works well to avoid position misalignments and acquires correct recognition despite less information on topological linkages.

With high spectral resolution hyperspectral imaging is capable of uncovering many subtle signal sources which cannot be known a priori or visually inspected. Such signal sources generally appear as anomalies in the data. Due to high correlation among spectral bands and sparsity of anomalies, a hyperspectral image can be e decomposed into two subspaces: a background subspace specified by a matrix with low rank dimensionality and an anomaly subspace specified by a sparse matrix with high rank dimensionality. This paper develops an approach to finding such low-high rank decomposition to identify anomaly subspace. Its idea is to formulate a convex constrained optimization problem that minimizes the nuclear norm of the background subspace and little ι 1 norm of the anomaly subspace subject to a decomposition of data space into background and anomaly subspaces. By virtue of such a background-anomaly decomposition the commonly used RX detector can be implemented in the sense that anomalies can be separated in the anomaly subspace specified by a sparse matrix. Experimental results demonstrate that the background-anomaly subspace decomposition can actually improve and enhance RXD performance.

Identification of materials from calibrated radiance data collected by an airborne imaging spectrometer depends strongly on the atmospheric and illumination conditions at the time of collection. This paper presents a methodology for identifying material spectra using the assumption that each unique material class forms a lower-dimensional manifold (surface) in the higher-dimensional spectral radiance space and that all image spectra reside on, or near, these theoretic manifolds. Using a physical model, a manifold characteristic of the target material exposed to varying illumination and atmospheric conditions is formed. A graph-based model is then applied to the radiance data to capture the intricate structure of each material manifold followed by the application of the commute time distance (CTD) transformation to separate the target manifold from the background. Detection algorithms are than applied in the CTD subspace. This nonlinear transformation is based on a Markov-chain model of a random walk on a graph and is derived from an eigendecomposition of the pseudoinverse of the graph Laplacian matrix. This paper discusses the properties of the CTDtransformation, the atmospheric and illumination parameters varied in the physics-based model and demonstrates the influence the target manifold samples have on the orientation of the coordinate axes in the transformed space. A comparison between detection performance in the CTD subspace and spectral radiance space is also given for two hyperspectral images.