Discriminative Random Fields Research Articles

Aspect Based Sentiment Analysis

Sentiment analysis, which addresses the computational treatment of opinion, sentiment, and subjectivity in text, has received considerable attention in recent years. In contrast to the traditional coarse-grained sentiment analysis tasks, such as document-level sentiment classification, we are interested in the fine-grained aspect-based sentiment analysis that aims to identify aspects that users comment on and these aspects’ polarities. Aspect-based sentiment analysis relies heavily on syntactic features. However, the reviews that this task focuses on are natural and spontaneous, thus posing a challenge to syntactic parsers. In this paper, we address this problem by proposing a framework of adding a sentiment sentence compression (Sent Comp) step before performing the aspect-based sentiment analysis. We apply a discriminative conditional random field model, with certain special features, to automatically compress sentiment sentences. Sentiment analysis is contextual mining of text which identifies and extracts subjective information in textual data. sentiment analysis proves to be an incredible asset for users to extract essential information and assists organizations with understanding the social sentiment of their brand, product or service while monitoring online conversations.

RF-HoDRF: High-Order Hybrid Discriminative Random Field Improved by Two-Layer Random Forest for SAR Image Change Detection

For better exploiting discriminative texture features and encoding high-level structures, this letter presents a high- order hybrid discriminative random field improved by two-layer random forest, abbreviated as RF-HoDRF, for synthetic aperture radar (SAR) image change detection. First, RF-HoDRF constructs a two-layer random forest (TL-RF) model to realize the selection of high-dimensional texture features, and then provides the class probabilities for constructing the unary potential in RF-HoDRF. Second, it defines a high-order potential on high-order cliques generated by superpixels to encode the high-level structures and maintain the region consistency. Finally, considering the pairwise potential by improved generalized Ising model and the statistics by generalized Gamma distribution (GΓD), the RF-HoDRF model is derived under the discriminative model framework. Then, by iteratively maximizing the local posterior probabilities, the class labels and the parameters are optimally estimated until they converge. Extensive comparisons and ablation experiments on measured SAR images verify the effectiveness of our method, and demonstrate that discriminative features selection and high-order structures maintenance have great contributions to improving change detection performances.

Sensor-based image manipulation localization with Discriminative Random fields and Graph Cut

We present PRNU-based image manipulation localization as a probabilistic labeling task in a flexible discriminative random field (DRF) setup. Instead of reaching local decisions independent of each other, discriminative random fields incorporate local inter-label dependencies while keeping the formulation general enough to make label assignments depend on both local and non-local image characteristics. With an improved form of association potential combining normalized correlation and the deviation of the measured correlation from the expected correlation and an interaction potential defined as the weighted L2 norm squared between intensities of neighboring pixels, we were able to localize even considerably small manipulations on realistic tampered images. We experimented with different combinations of window sizes to capture features to predict the correlation more accurately than already existing algorithms. Experimental results indicate that our algorithm outperforms recent state of the art methods based on multiscale analysis strategies. We also found that for inspecting manipulated images which are JPEG compressed, it helps to train the predictor with JPEG images rather than with uncompressed images and for all quality factors, it is possible to work with two predictors, one trained for images with lower quality factors and another for higher quality factors.

Nonstationary PolSAR Image Classification by Deep-Features-Based High-Order Triple Discriminative Random Field

Aiming at exploiting the discriminative deep features and encoding the high-level structures, this letter presents a deep-features-based high-order triple discriminative random field model, abbreviated as DF-HoTDF, for nonstationary polarimetric synthetic aperture radar (PolSAR) image classification. First, the DF-HoTDF model extracts the discriminative deep features by a graph-based complex-valued 3-D convolutional neural network (CV-3-D-CNN) and then constructs the unary potential by a negative log function. Second, it introduces an auxiliary field u to explicitly regulate the nonstationary label patterns of the PolSAR image and then constructs a pairwise potential guided by u to capture greater pairwise label interactions. Third, it defines a high-order potential on high-order cliques to encode high-level structures. Finally, under the discriminative model framework, the DF-HoTDF model has a weighted fusion of the unary potential, the pairwise potential, and the high-order potential. Then, with the DF-HoTDF model, we iteratively optimize the class label and the stationary maps until they converge. The experimental results demonstrate that the proposed DF-HoTDF model is of superior performances in nonstationary PolSAR image classification and that it can provide better label consistency in homogeneous region and better target structures and edge locations in heterogeneous region.

Semisupervised Discriminative Random Field for Hyperspectral Image Classification

The integration of spectral and spatial information is crucial in remotely sensed hyperspectral image classification. Some available approaches extract spatial features before classification, while other techniques include spatial information as a spatial regularizer. Due to the model complexity, these methods generally assume that a given pixel is only dependent on itself and its neighboring pixels. However, it is known that there are strong interactions among the pixels in different areas of the image, and as a result, this general assumption can limit the classifier performance. In this work, we develop a new semi-supervised discriminative random field (DRF) technique for spatial-spectral classification of hyperspectral images, which constructs the complete model using the whole image. Furthermore, to cope with the unlabeled samples in the model, a new expectation maximization (EM) based algorithm is introduced to optimize the learning of DRF parameters, where the missing labels of these samples serve as the latent variable. As compared to other conventional learning approaches such as maximum a posteriori marginal (MPM) labeling, the proposed approach can obtain better optimization that leads to improved classification performance. Our experimental results, conducted using real hyperspectral datasets, indicate that the newly developed DRF algorithm exhibits very good performance.

Composite Kernel and Hybrid Discriminative Random Field Model Based on Feature Fusion for PolSAR Image Classification

To effectively fuse the high-dimensional features, in this letter, we propose the composite kernel and hybrid discriminative random field model, abbreviated as CK-hybrid discriminative random field (HDRF), for polarimetric synthetic aperture radar (PolSAR) image classification. In the CK-HDRF model, given high-dimensional features with different characteristics, the unary potential is constructed by relating multiple kernel k-means (MKKM) clustering to the traditional HDRF model. In this way, the high-dimensional decomposition and texture features can be well fused, thus making their deserved contributions to the inference of the attributive class and further increasing the discrimination capacity of CK-HDRF. The pairwise potential is constructed by the generalized Ising model with an additional edge penalty function, and thus, it can well capture the underlying spatial relationship and maintain the edge locations in classification. Moreover, the statistics of PolSAR data are modeled by the Wishart-generalized gamma (WG Γ) distribution. Experiments on real PolSAR images demonstrate the effectiveness of CK-HDRF in classification.

Subsurface Source Zone Characterization and Uncertainty Quantification Using Discriminative Random Fields

AbstractA novel statistical approach is developed and implemented for the stochastic reconstruction of nonaqueous phase liquid (NAPL) source zone realizations and the quantification of source zone metrics and associated uncertainty. The approach employs discriminative random field (DRF) models, to simulate the spatial distributions and relationships among source zone properties (i.e., permeability, NAPL saturation, and aqueous concentration distributions) consistent with commonly collected field data. Application of DRF models requires a limited number of full‐scale simulations to train the model parameters. Monte Carlo sampling methods based on these trained models then provide an efficient method to generate contaminant mass realizations conditioned on measured boreholes, bypassing the need to run computationally intensive, partial differential equation‐based simulations of physical flow and transport. Postprocessing of these realizations yields approximations of uncertainty to inform further sampling for characterization and remediation. The reconstructed contaminant mass realizations provide sufficient information for calculating averaged characterization metrics, such as total contaminant mass and pool fraction, used to predict source zone longevity, mass recovery behavior, and remedial performance. The model performance is evaluated through comparisons of these predicted source zone metrics with those obtained from the “true” mass distributions generated with validated flow and transport models. These comparisons clearly demonstrate that stochastic application of a DRF model can reconstruct realistic saturation and concentration fields, conditioned to borehole data at different times. The present study should be viewed as the first step in generating a three‐dimensional characterization tool that can be applied over a wide range of conditions observed at contaminated sites.

Fuzziness Modeling of Polarized Scattering Mechanisms and PolSAR Image Classification Using Fuzzy Triplet Discriminative Random Fields

Dominant scattering mechanism (DSM) obtained by Freeman decomposition is significant for polarimetric synthetic aperture radar (PolSAR) image classification. To preserve the purity of scattering characteristics, it restricts pixels in a scattering category to be classified with other pixels in the same scattering category. However, due to the speckle and the limited image resolution, it is difficult to obtain the DSMs of some pixels, which are defined as the fuzziness of polarized scattering mechanisms. Therefore, we first consider a particular—and pertinent—auxiliary field, and then propose the fuzzy triplet discriminative random fields (FTDF) model to describe the fuzziness of polarized scattering mechanisms, thus categorizing the scattering mechanisms into four classes: surface scattering, double-bounce scattering, volume scattering, and mixed scattering. The pixels in the first three categories are with specific DSMs, and the FTDF model introduces an exponential kernel distance to combine the multiple features of PolSAR data into classification. For the pixels in the mixed scattering, FTDF introduces a fuzzy clustering algorithm regularized by Kullback–Leibler information to consider the fuzzy DSMs, thus enhancing the classification. Then the fuzziness modeling of polarized scattering mechanisms can guide the classification of PolSAR images. The experimental results on real PolSAR images demonstrate the effectiveness of the FTDF model, and illustrate that it can improve the classification accuracy, and simultaneously preserve the purity of scattering mechanisms.

Spatial Prior Fuzziness Pool-Based Interactive Classification of Hyperspectral Images

Acquisition of labeled data for supervised Hyperspectral Image (HSI) classification is expensive in terms of both time and costs. Moreover, manual selection and labeling are often subjective and tend to induce redundancy into the classifier. Active learning (AL) can be a suitable approach for HSI classification as it integrates data acquisition to the classifier design by ranking the unlabeled data to provide advice for the next query that has the highest training utility. However, multiclass AL techniques tend to include redundant samples into the classifier to some extent. This paper addresses such a problem by introducing an AL pipeline which preserves the most representative and spatially heterogeneous samples. The adopted strategy for sample selection utilizes fuzziness to assess the mapping between actual output and the approximated a-posteriori probabilities, computed by a marginal probability distribution based on discriminative random fields. The samples selected in each iteration are then provided to the spectral angle mapper-based objective function to reduce the inter-class redundancy. Experiments on five HSI benchmark datasets confirmed that the proposed Fuzziness and Spectral Angle Mapper (FSAM)-AL pipeline presents competitive results compared to the state-of-the-art sample selection techniques, leading to lower computational requirements.

Superpixel-Based Hybrid Discriminative Random Field for Fast PolSAR Image Classification

Performance of the powerful discriminative random field (DRF) model for image processing and analysis is easily affected by the inherent speckle noise and the time-consuming iteration. Therefore, in this paper, a superpixel-based hybrid DRF ( sp -HDRF) model is proposed for fast polarimetric synthetic aperture radar (PolSAR) image classification. The sp -HDRF model realizes the classification by two steps. First, the simple linear iterative clustering algorithm, which is modified by introducing the ratio of exponentially weighted averages operator, is utilized to obtain a superpixel graph with more accurate edge locations. Second, the conditional posterior distribution and the inference formula of the sp -HDRF model are derived on the superpixel graph, which is a generalization of a DRF model on the pixel. Then, the sp -HDRF model is applied to implement classification. Finally, the sp -HDRF model has the fusion of the polarimetric scattering features, the statistics, and the spatial relationships of the image. The experimental results on the real PolSAR images demonstrate the effectiveness of the sp -HDRF model and illustrate that it can provide stronger noise immunity, obtain smoother homogeneous areas in classification, and enhance the computational efficiency simultaneously.

Triplet Hybrid Discriminative Fields Model Based on Bayesian Fusion for Unsupervised Nonstationary SAR Image Multiclass Segmentation

The discriminative random fields (DRF) model is suitable for analyzing images with complex textural structures and has achieved promising results in image segmentation. However, the DRF model does not consider the nonstationarity of synthetic aperture radar (SAR) images and lacks the ability to model SAR scattering statistics in nonstationary SAR image segmentation. In this paper, we propose a triplet hybrid discriminative random fields (THDF) model based on Bayesian fusion. According to its semantic structure, the THDF model belongs to hybrid discriminative models, and it provides the following promising contributions to nonstationary SAR image segmentation while inheriting the advantages of the discriminative models: first, it takes the nonstationarity of SAR images into account from the perspective of their texton appearances, and thus regulates the local label interaction patterns and considers the distribution differences of the congeneric image features in different stationary parts; and second, for nonstationary SAR images, it performs a fusion-type treatment of the nonstationary textural features and the SAR scattering statistics based on Bayesian fusion and, thus, captures the nonstationary information from SAR data in a more complete manner. The effectiveness of the proposed model is demonstrated through applications to both synthetic images and real SAR image segmentations.

PRNU-based Image Manipulation Localization with Discriminative Random Fields

PRNU-based Image Manipulation Localization with Discriminative Random Fields

Dempster–Shafer Fusion of Evidential Pairwise Markov Chains

Hidden Markov models have been extended in many directions, leading to pairwise Markov models, triplet Markov models, or discriminative random fields, all of which have been successfully applied in many fields covering signal and image processing. The Dempster–Shafer theory of evidence has also shown its interest in a wide range of situations involving reasoning under uncertainty and/or information fusion. There are, however, only few works dealing with both of these modeling tools simultaneously. The aim of this paper, which falls under this category of works, is to propose a general evidential Markov model offering wide modeling and processing possibilities regarding information imprecision, sensor unreliability, and data fusion. The main interest of the proposed model relies in the possibility of achieving, easily, the Dempster–Shafer fusion without destroying the Markovianity.

Discriminative Random Fields Based on Maximum Entropy Principle for Semisupervised SAR Image Change Detection

This paper proposes a novel semisupervised SAR images change detection algorithm using discriminative random fields based on maximum entropy principle (MEDRF). MEDRF is a discriminative model fused by two generative models, named as the bias model and the correction model, based on maximum entropy (ME) principle. In MEDRF model, we construct the bias model and the correction model on labeled samples and unlabeled samples, respectively, based on Markov random fields (MRF) to capture the multitemporal image information. Then, we deduce two constraints from the two generative models, and thus fuse the bias model and the correction model to derive MEDRF model according to ME principle subjected to the two constraints. In this way, the proposed MEDRF takes full advantages of the image information from the labeled samples and the unlabeled samples, especially including the spatial-contextual information, to provide an appropriate class boundary. In the experiment, we analyze the influence of the number of labeled samples to the performance of MEDRF model in semisupervised change detection to illustrate that MEDRF can achieve appropriate detection results even using a small number of labeled samples, and the experimental results on real SAR data demonstrate MEDRF model is able to achieve improvement in change detection over several methods proposed recently.

Graphics processing unit–accelerated computation of the Markov random fields and loopy belief propagation algorithms for hyperspectral image classification

We propose a commodity graphics processing units (GPUs)–based massively parallel efficient computation for spectral-spatial classification of hyperspectral images. The spectral-spatial classification framework is based on the marginal probability distribution which uses all of the information in the hyperspectral data. In this framework, first, the posterior class probability is modeled with discriminative random field in which the association potential is linked with a multinomial logistic regression (MLR) classifier and the interaction potential modeling the spatial information is linked to a Markov random field multilevel logistic (MLL) prior. Second, the maximizers of the posterior marginals are computed via the loopy belief propagation (LBP) method. In addition, the regressors of the multinominal logistic regression classifier are inferred by the logistic regression via variable splitting and augmented Lagrangian (LORSAL) algorithm. Although the spectral-spatial classification framework exhibited state-of-the-art accuracy performance with regard to similar approaches, its computational complexity is very high. We take advantage of the massively parallel computing capability of NVIDIA Tesla C2075 with the compute unified device architecture including a set of GPU-accelerated linear algebra libraries (CULA) to dramatically improve the computation speed of this hyperspectral image classification framework. The shared memory and the asynchronous transfer techniques are also used for further computationally efficient optimization. Real hyperspectral data sets collected by the National Aeronautics and Space Administration’s airborne visible infrared imaging spectrometer and the reflective optics system imaging spectrometer system are used for effectiveness evaluation. The results show that we achieved a speedup of 92-fold on LORSAL, 69-fold on MLR, 127-fold on MLL, 160-fold on LBP, and 73-fold on the whole spectral-spatial classification framework as compared with its single-core central processing unit counterpart, respectively.

Detection of Building in Natural Images with one New Discriminative Random Fields

This paper presents a new Discriminative Random Fields (DRFs) framework. Based on the DRFs framework proposed by Kumar and Hebert, the following improvements have been conducted. Firstly, the interaction potential and the associated potential model are simplified. Secondly, we reduce the dimension of the multi-scale features, re-definedimension of the single-scale feature, and increase the color feature of Building. Thirdly,the quasi-Newton method with linear search and gradient descent method are adopted to solve parameters, whichget a simple model and achieve good performance. Finally, the partition function of the DRF is eliminatedby using Pseudo-likelihood method for parameter learning. The simulation results show thatthe proposed method’s false positive rate is lower than the method from Kumar and Hebert, while the correct rate and detection ratearehigher than their experimental effects after these improvements.

Max‐margin discriminative random fields for multimodal human action recognition

Proposed is a max-margin discriminative random fields model for multimodal human action recognition. To incorporate multiple modalities for joint modelling, a specific graphical structure with parallel sequential observations and related hidden-state layers is designed. Moreover, the corresponding potential functions for model formulation are designed. For model learning, the max-margin learning method is proposed to discover both latent correlation among multimodal data and temporal context within individual modality. A comparison experiment shows that the proposed model can boost the performance of human action recognition by taking advantage of complementary characteristics from multiple modalities.

ECOC-DRF: Discriminative random fields based on error correcting output codes

We present ECOC-DRF, a framework where potential functions for Discriminative Random Fields are formulated as an ensemble of classifiers. We introduce the label trick, a technique to express transitions in the pairwise potential as meta-classes. This allows to independently learn any possible transition between labels without assuming any pre-defined model. The Error Correcting Output Codes matrix is used as ensemble framework for the combination of margin classifiers. We apply ECOC-DRF to a large set of classification problems, covering synthetic, natural and medical images for binary and multi-class cases, outperforming state-of-the art in almost all the experiments.

Automatic knee cartilage segmentation from multi-contrast MR images using support vector machine classification with spatial dependencies

Accurate segmentation of knee cartilage is required to obtain quantitative cartilage measurements, which is crucial for the assessment of knee pathology caused by musculoskeletal diseases or sudden injuries. This paper presents an automatic knee cartilage segmentation technique which exploits a rich set of image features from multi-contrast magnetic resonance (MR) images and the spatial dependencies between neighbouring voxels. The image features and the spatial dependencies are modelled into a support vector machine (SVM)-based association potential and a discriminative random field (DRF)-based interaction potential. Subsequently, both potentials are incorporated into an inference graphical model such that the knee cartilage segmentation is cast into an optimal labelling problem which can be efficiently solved by loopy belief propagation. The effectiveness of the proposed technique is validated on a database of multi-contrast MR images. The experimental results show that using diverse forms of image and anatomical structure information as the features are helpful in improving the segmentation, and the joint SVM-DRF model is superior to the classification models based solely on DRF or SVM in terms of accuracy when the same features are used. The developed segmentation technique achieves good performance compared with gold standard segmentations and obtained higher average DSC values than the state-of-the-art automatic cartilage segmentation studies.

The unified extreme learning machines and discriminative random fields for automatic knee cartilage and meniscus segmentation from multi-contrast MR images

Segmenting articular cartilage and meniscus from magnetic resonance (MR) images is an essential task for the assessment of knee pathology. Most of the previous classification-based works for cartilage and meniscus segmentation only rely on independent labellings by a classifier, but do not consider the spatial context interaction. The labels of most image voxels are actually dependent upon their neighbours. In this study, we present an automatic knee segmentation system working on multi-contrast MR images where a novel classification model unifying an extreme learning machine (ELM)-based association potential and a discriminative random field (DRF)-based interaction potential is proposed. The DRF model introduces spatial dependencies between neighbouring voxels to the independent ELM classification. We exploit a rich set of features From multi-contrast MR images to train the proposed classification model and perform the loopy belief propagation for the inference. The proposed model is evaluated on multi-contrast MR datasets acquired from 11 subjects with results outperforming the independent classifiers in terms of segmentation accuracy of both cartilages and menisci.