The Experts below are selected from a list of 144804 Experts worldwide ranked by ideXlab platform
Xiaoyang Tan - One of the best experts on this subject based on the ideXlab platform.
-
Sparsity preserving discriminant analysis for single Training Image face recognition
Pattern Recognition Letters, 2010Co-Authors: Lishan Qiao, Songcan Chen, Xiaoyang TanAbstract:Single Training Image face recognition is one of the main challenges to appearance-based pattern recognition techniques. Many classical dimensionality reduction methods such as LDA have achieved success in face recognition field, but cannot be directly used to the single Training Image scenario. Recent graph-based semi-supervised dimensionality reduction (SSDR) provides a feasible strategy to deal with such problem. However, most of the existing SSDR algorithms such as semi-supervised discriminant analysis (SDA) are locality-oriented and generally suffer from the following issues: (1) they need a large number of unlabeled Training samples to estimate the manifold structure in data, but such extra samples may not be easily obtained in a given face recognition task; (2) they model the local geometry of data by the nearest neighbor criterion which generally fails to obtain sufficient discriminative information due to the high-dimensionality of face Image space; (3) they construct the underlying adjacency graph (or data-dependent regularizer) using a fixed neighborhood size for all the sample points without considering the actual data distribution. In this paper, we develop a new graph-based SSDR algorithm called sparsity preserving discriminant analysis (SPDA) to address these problems. More specifically, (1) the graph in SPDA is constructed by sparse representation, and thus the local structure in data is automatically modeled instead of being manually predefined. (2) With the natural discriminative power of sparse representation, SPDA can remarkably improve recognition performance only resorting to very few extra unlabeled samples. (3) A simple ensemble strategy is developed to accelerate graph construction, which results in an efficient ensemble SPDA algorithm. Extensive experiments on both toy and real face data sets are provided to validate the feasibility and effectiveness of the proposed algorithm.
-
Recognizing partially occluded, expression variant faces from single Training Image per person with SOM and soft k-NN ensemble
IEEE transactions on neural networks, 2005Co-Authors: Xiaoyang Tan, Zhihua Zhou, Songcan Chen, Fuyan ZhangAbstract:Most classical template-based frontal face recognition techniques assume that multiple Images per person are available for Training, while in many real-world applications only one Training Image per person is available and the test Images may be partially occluded or may vary in expressions. This paper addresses those problems by extending a previous local probabilistic approach presented by Martinez, using the self-organizing map (SOM) instead of a mixture of Gaussians to learn the subspace that represented each individual. Based on the localization of the Training Images, two strategies of learning the SOM topological space are proposed, namely to train a single SOM map for all the samples and to train a separate SOM map for each class, respectively. A soft k nearest neighbor (soft k-NN) ensemble method, which can effectively exploit the outputs of the SOM topological space, is also proposed to identify the unlabeled subjects. Experiments show that the proposed method exhibits high robust performance against the partial occlusions and variant expressions.
Songcan Chen - One of the best experts on this subject based on the ideXlab platform.
-
Sparsity preserving discriminant analysis for single Training Image face recognition
Pattern Recognition Letters, 2010Co-Authors: Lishan Qiao, Songcan Chen, Xiaoyang TanAbstract:Single Training Image face recognition is one of the main challenges to appearance-based pattern recognition techniques. Many classical dimensionality reduction methods such as LDA have achieved success in face recognition field, but cannot be directly used to the single Training Image scenario. Recent graph-based semi-supervised dimensionality reduction (SSDR) provides a feasible strategy to deal with such problem. However, most of the existing SSDR algorithms such as semi-supervised discriminant analysis (SDA) are locality-oriented and generally suffer from the following issues: (1) they need a large number of unlabeled Training samples to estimate the manifold structure in data, but such extra samples may not be easily obtained in a given face recognition task; (2) they model the local geometry of data by the nearest neighbor criterion which generally fails to obtain sufficient discriminative information due to the high-dimensionality of face Image space; (3) they construct the underlying adjacency graph (or data-dependent regularizer) using a fixed neighborhood size for all the sample points without considering the actual data distribution. In this paper, we develop a new graph-based SSDR algorithm called sparsity preserving discriminant analysis (SPDA) to address these problems. More specifically, (1) the graph in SPDA is constructed by sparse representation, and thus the local structure in data is automatically modeled instead of being manually predefined. (2) With the natural discriminative power of sparse representation, SPDA can remarkably improve recognition performance only resorting to very few extra unlabeled samples. (3) A simple ensemble strategy is developed to accelerate graph construction, which results in an efficient ensemble SPDA algorithm. Extensive experiments on both toy and real face data sets are provided to validate the feasibility and effectiveness of the proposed algorithm.
-
Recognizing partially occluded, expression variant faces from single Training Image per person with SOM and soft k-NN ensemble
IEEE transactions on neural networks, 2005Co-Authors: Xiaoyang Tan, Zhihua Zhou, Songcan Chen, Fuyan ZhangAbstract:Most classical template-based frontal face recognition techniques assume that multiple Images per person are available for Training, while in many real-world applications only one Training Image per person is available and the test Images may be partially occluded or may vary in expressions. This paper addresses those problems by extending a previous local probabilistic approach presented by Martinez, using the self-organizing map (SOM) instead of a mixture of Gaussians to learn the subspace that represented each individual. Based on the localization of the Training Images, two strategies of learning the SOM topological space are proposed, namely to train a single SOM map for all the samples and to train a separate SOM map for each class, respectively. A soft k nearest neighbor (soft k-NN) ensemble method, which can effectively exploit the outputs of the SOM topological space, is also proposed to identify the unlabeled subjects. Experiments show that the proposed method exhibits high robust performance against the partial occlusions and variant expressions.
-
enhanced pc 2 a for face recognition with one Training Image per person
Pattern Recognition Letters, 2004Co-Authors: Songcan Chen, Daoqiang Zhang, Zhihua ZhouAbstract:Abstract Recently, a method called (PC) 2 A was proposed to deal with face recognition with one Training Image per person. As an extension of the standard eigenface technique, (PC) 2 A combines linearly each original face Image with its corresponding first-order projection into a new face and then performs principal component analysis (PCA) on a set of the newly combined (Training) Images. It was reported that (PC) 2 A could achieve higher accuracy than the eigenface technique through using 10–15% fewer eigenfaces. In this paper, we generalize and further enhance (PC) 2 A along two directions. In the first direction, we combine the original Image with its second-order projections as well as its first-order projection in order to acquire more information from the original face, and then similarly apply PCA to such a set of the combined Images. In the second direction, instead of combining them, we still regard the projections of each original Image as single derived Images to augment Training Image set, and then perform PCA on all the Training Images available, including the original ones and the derived ones. Experiments on the well-known FERET database show that the enhanced versions of (PC) 2 A are about 1.6–3.5% more accurate and use about 47.5–64.8% fewer eigenfaces than (PC) 2 A.
Tanxiaoyang - One of the best experts on this subject based on the ideXlab platform.
-
Sparsity preserving discriminant analysis for single Training Image face recognition
Pattern Recognition Letters, 2010Co-Authors: Qiaolishan, Chensongcan, TanxiaoyangAbstract:Single Training Image face recognition is one of the main challenges to appearance-based pattern recognition techniques. Many classical dimensionality reduction methods such as LDA have achieved su...
-
Recognizing partially occluded, expression variant faces from single Training Image per person with SOM and soft k-NN ensemble
IEEE Transactions on Neural Networks, 2005Co-Authors: Tanxiaoyang, Zhouzhi-hua, Chensongcan, ZhangfuyanAbstract:Most classical template-based frontal face recognition techniques assume that multiple Images per person are available for Training, while in many real-world applications only one Training Image pe...
Jef Caers - One of the best experts on this subject based on the ideXlab platform.
-
uncertainty in Training Image based inversion of hydraulic head data constrained to ert data workflow and case study
Water Resources Research, 2015Co-Authors: Thomas Hermans, Frederic Nguyen, Jef CaersAbstract:In inverse problems, investigating uncertainty in the posterior distribution of model parameters is as important as matching data. In recent years, most efforts have focused on techniques to sample the posterior distribution with reasonable computational costs. Within a Bayesian context, this posterior depends on the prior distribution. However, most of the studies ignore modeling the prior with realistic geological uncertainty. In this paper, we propose a workflow inspired by a Popper-Bayes philosophy that data should first be used to falsify models, then only be considered for matching. We propose a workflow consisting of three steps: (1) in defining the prior, we interpret multiple alternative geological scenarios from literature (architecture of facies) and site-specific data (proportions of facies). Prior spatial uncertainty is modeled using multiple-point geostatistics, where each scenario is defined using a Training Image. (2) We validate these prior geological scenarios by simulating electrical resistivity tomography (ERT) data on realizations of each scenario and comparing them to field ERT in a lower dimensional space. In this second step, the idea is to probabilistically falsify scenarios with ERT, meaning that scenarios which are incompatible receive an updated probability of zero while compatible scenarios receive a nonzero updated belief. (3) We constrain the hydrogeological model with hydraulic head and ERT using a stochastic search method. The workflow is applied to a synthetic and a field case studies in an alluvial aquifer. This study highlights the importance of considering and estimating prior uncertainty (without data) through a process of probabilistic falsification.
-
Multiple-Point Geostatistics: Stochastic Modeling with Training Images - Universal kriging with Training Images
Spatial Statistics, 2015Co-Authors: Thomas Romary, Jef CaersAbstract:Abstract In the past decade, the Training Image (TI) has received considerable attention as a source for modeling spatial continuity in geostatistics. In this paper, the use of TIs in the context of kriging is investigated, specifically universal kriging (UK). Traditionally, kriging relies on a random function model formulation whereby the target variable is decomposed into a trend and residual. While the theory is firm and elegant, the actual practice of UK remains challenging; in particular when data is sparse, and the modeler has to decide what to model as the trend and as the residual. This paper juxtaposes this variogram-based universal kriging (UK-v) with a TI-based approach (UK-TI). It is found that the latter need not rely on random function theory, but rather on the specification of a TI on which “universal” conditions are verified. Through illustrations with examples, it is seen that the modeling challenge in UK-TI is on the Training Image. Using a Monte Carlo study, the statistical performance of both methods is found to be comparable. Recommendations on which method to choose, based on practical criteria, are also formulated. Additionally, the study provides more insight into the use of the TI in general, including in multiple-point geostatistics.
-
Assessing the Probability of Training Image-Based Geological Scenarios Using Geophysical Data
Lecture Notes in Earth System Sciences, 2013Co-Authors: Thomas Hermans, Jef Caers, Frederic NguyenAbstract:The construction of Training Images (TIs) depicting the geological prior is one of the most critical step in multiple-point statistics. Geophysical techniques may be used to reduce the uncertainty in the understanding of prior geological scenarios. We developed a methodology to verify the consistency of geophysical data with independently-built TIs. Synthetic geophysical models built from TI scenarios are compared, using multidimensional scaling, with inverted models from field surveys to check if TIs are consistent with geophysical models. Then, the probability of each TI scenario is computed. A cluster analysis enables to determine which parameters used in building the TIs are most impacting the geophysical response. The methodology is tested using ERT to analyze TI scenarios in the Meuse River alluvial aquifer (Belgium)
-
Training Image based scenario modeling of fractured reservoirs for flow uncertainty quantification
Computational Geosciences, 2013Co-Authors: Andre Jung, Darryl Fenwick, Jef CaersAbstract:Geological characterization of naturally fractured reservoirs is potentially associated with large uncertainty. However, the geological modeling of discrete fracture networks (DFN) is considerably disconnected from uncertainty modeling based on conventional flow simulators in practice. DFN models provide a geologically consistent way of modeling fractures in reservoirs. However, flow simulation of DFN models is currently infeasible at the field scale. To translate DFN models to dual media descriptions efficiently and rapidly, we propose a geostatistical approach based on patterns. We will use experimental design to capture the uncertainties in the fracture description and generate DFN models. The DFN models are then upscaled to equivalent continuum models. Patterns obtained from the upscaled DFN models are reduced to a manageable set and used as Training Images for multiple-point statistics (MPS). Once the Training Images are obtained, they allow for fast realization of dual-porosity descriptions with MPS directly, while circumventing the time-consuming process of DFN modeling and upscaling. We demonstrate our ideas on a realistic Middle East-type fractured reservoir system.
-
Comparing Training-Image Based Algorithms Using an Analysis of Distance
Mathematical Geosciences, 2013Co-Authors: Xiaojin Tan, Pejman Tahmasebi, Jef CaersAbstract:As additional multiple-point statistical (MPS) algorithms are developed, there is an increased need for scientific ways for comparison beyond the usual visual comparison or simple metrics, such as connectivity measures. In this paper, we start from the general observation that any (not just MPS) geostatistical simulation algorithm represents two types of variability: (1) the within-realization variability, namely, that realizations reproduce a spatial continuity model (variogram, Boolean, or Training-Image based), (2) the between-realization variability representing a model of spatial uncertainty. In this paper, it is argued that any comparison of algorithms needs, at a minimum, to be based on these two randomizations. In fact, for certain MPS algorithms, it is illustrated with different examples that there is often a trade-off: Increased pattern reproduction entails reduced spatial uncertainty. In this paper, the subjective choice that the best algorithm maximizes pattern reproduction is made while at the same time maximizes spatial uncertainty. The discussion is also limited to fairly standard multiple-point algorithms and that our method does not necessarily apply to more recent or possibly future developments. In order to render these fundamental principles quantitative, this paper relies on a distance-based measure for both within-realization variability (pattern reproduction) and between-realization variability (spatial uncertainty). It is illustrated in this paper that this method is efficient and effective for two-dimensional, three-dimensional, continuous, and discrete Training Images.
Gregoire Mariethoz - One of the best experts on this subject based on the ideXlab platform.
-
Downscaling Images with Trends Using Multiple-Point Statistics Simulation: An Application to Digital Elevation Models
Mathematical Geosciences, 2019Co-Authors: Luiz Gustavo Rasera, Mathieu Gravey, Stuart N. Lane, Gregoire MariethozAbstract:Remote sensing and geophysical imaging techniques are often limited in terms of spatial resolution. This prevents the characterization of physical properties and processes at scales finer than the spatial resolution provided by the imaging sensor. In the last decade, multiple-point statistics simulation has been successfully used for downscaling problems. In this approach, the missing fine-scale structures are imported from a Training Image which describes the correspondence between coarse and equivalent fine-scale structures. However, in many cases, large variations in the amplitude of the Imaged physical attribute, known as trends, pose a challenge for the detection and simulation of these fine-scale features. Here, we develop a novel multiple-point statistics simulation method for downscaling coarse-resolution Images with trends. The proposed algorithm relies on a multi-scale sequential simulation framework. Trends in the data are handled by an inbuilt decomposition of the target variable into a deterministic trend component and a stochastic residual component at multiple scales. We also introduce the application of kernel weighting for computing distances between data events and probability aggregation operations for integrating different support data based on a distance-to-probability transformation function. The algorithm is benchmarked against two-point and multiple-point statistics simulation methods, and a deterministic interpolation technique. Results show that the approach is able to cope with non-stationary data sets and scenarios in which the statistics of the Training Image differ from the conditioning data statistics. Two case studies using digital elevation models of mountain ranges in Switzerland illustrate the method.
-
Efficient Training Image selection for multiple-point geostatistics via analysis of contours
Computers & Geosciences, 2019Co-Authors: Mohammad Javad Abdollahifard, Mohammad Baharvand, Gregoire MariethozAbstract:Abstract Multiple-point statistics (MPS) methods have emerged as efficient tools for environmental modelling, however their efficiency highly depends on the availability of appropriate Training Images (TIs). We introduce an efficient method for selecting one compatible TI among a proposed set, based on a measure of compatibility with available conditioning data. While existing approaches to do this consider all available data-events in the simulation grid, we concentrate on a limited number of data-events around the contours and edges of the Image. The proposed method is evaluated with different sampling rates, based on hundreds of sample sets extracted from binary, categorical and continuous Images, and compared with exhaustive data-event extraction. Our experiments show that the proposed method improves the required CPU-time by up to two orders of magnitude and at the same time leads to a slight improvement in the recognition accuracy.
-
Improving in situ data acquisition using Training Images and a Bayesian mixture model
Computers & Geosciences, 2016Co-Authors: Mohammad Javad Abdollahifard, Gregoire Mariethoz, Mohammadreza PourfardAbstract:Estimating the spatial distribution of physical processes using a minimum number of samples is of vital importance in earth science applications where sampling is costly. In recent years, Training Image-based methods have received a lot of attention for interpolation and simulation. However, Training Images have never been employed to optimize spatial sampling process. In this paper, a sequential compressive sampling method is presented which decides the location of new samples based on a Training Image. First, a Bayesian mixture model is developed based on the Training patterns. Then, using this model, unknown values are estimated based on a limited number of random samples. Since the model is probabilistic, it allows estimating local uncertainty conditionally to the available samples. Based on this, new samples are sequentially extracted from the locations with maximum uncertainty. Experiments show that compared to a random sampling strategy, the proposed supervised sampling method significantly reduces the number of samples needed to achieve the same level of accuracy, even when the Training Image is not optimally chosen. The method has the potential to reduce the number of observations necessary for the characterization of environmental processes. HighlightsA Training Image-based method is developed for sampling design.The method has a sequential nature.New samples are extracted from highly uncertain areas.A Bayesian mixture model is employed for estimating uncertainty values.The method significantly reduced the number of samples required for reconstruction of a field.
-
Integrating multiple scales of hydraulic conductivity measurements in Training Image-based stochastic models
Water Resources Research, 2015Co-Authors: K. Mahmud, Gregoire Mariethoz, Andy Baker, Ashish SharmaAbstract:Hydraulic conductivity is one of the most critical and at the same time one of the most uncertain parameters in many groundwater models. One problem commonly faced is that the data are usually not collected at the same scale as the discretized elements used in a numerical model. Moreover, it is common that different types of hydraulic conductivity measurements, corresponding to different spatial scales, coexist in a studied domain, which have to be integrated simultaneously. Here we address this issue in the context of Image Quilting, one of the recently developed multiple-point geostatistics methods. Based on a Training Image that represents fine-scale spatial variability, we use the simplified renormalization upscaling method to obtain a series of upscaled Training Images that correspond to the different scales at which measurements are available. We then apply Image Quilting with such a multiscale Training Image to be able to incorporate simultaneously conditioning data at several spatial scales of heterogeneity. The realizations obtained satisfy the conditioning data exactly across all scales, but it can come at the expense of a small approximation in the representation of the physical scale relationships. In order to mitigate this approximation, we iteratively apply a kriging-based correction to the finest scale that ensures local conditioning at the coarsest scales. The method is tested on a series of synthetic examples where it gives good results and shows potential for the integration of different measurement methods in real-case hydrogeological models.
-
verifying the high order consistency of Training Images with data for multiple point geostatistics
Computers & Geosciences, 2014Co-Authors: Cristian Perez, Gregoire Mariethoz, Julian M OrtizAbstract:Parameter inference is a key aspect of spatial modeling. A major appeal of variograms is that they allow inferring the spatial structure solely based on conditioning data. This is very convenient when the modeler does not have a ready-made geological interpretation. To date, such an easy and automated interpretation is not available in the context of most multiple-point geostatistics applications. Because Training Images are generally conceptual models, their preparation is often based on subjective criteria of the modeling expert. As a consequence, selection of an appropriate Training Image is one of the main issues one must face when using multiple-point simulation. This paper addresses the development of a geostatistical tool that addresses two separate problems. It allows (1) ranking Training Images according to their relative compatibility to the data, and (2) obtaining an absolute measure quantifying the consistency between Training Image and data in terms of spatial structure. For both, two alternative implementations are developed. The first one computes the frequency of each pattern in each Training Image. This method is statistically sound but computationally demanding. The second implementation obtains similar results at a lesser computational cost using a direct sampling approach. The applicability of the methodologies is successfully evaluated in two synthetic 2D examples and one real 3D mining example at the Escondida Norte deposit. The method allows ranking TI?s based on their relative high-order consistency with data.The method also allows obtaining an absolute consistency measure.The CPU time scales considerably down when a direct sampling approach is used.Applicability is successfully evaluated in different real and synthetic case studies.FORTRAN code of the program is presented and a practical example is attached.
