Robust Latent Subspace Learning for Image Classification

Orthogonal Subspace Representation for Generative Adversarial Networks

Disentanglement learning aims to separate explanatory factors of variation so that different attributes of the data can be well characterized and isolated, which promotes efficient inference for downstream tasks. Mainstream disentanglement approaches based on generative adversarial networks (GANs) learn interpretable data representation. However, most typical GAN-based works lack the discussion of the latent subspace, causing insufficient consideration of the variation of independent factors. Although some recent research analyzes the latent space on pretrained GANs for image editing, they do not emphasize learning representation directly from the subspace perspective. Appropriate subspace properties could facilitate corresponding feature representation learning to satisfy the independent variation requirements of the obtained explanatory factors, which is crucial for better disentanglement. In this work, we propose a unified framework for ensuring disentanglement, which fully investigates latent subspace learning (SL) in GAN. The novel GAN-based architecture explores orthogonal subspace representation (OSR) on vanilla GAN, named OSRGAN. To guide a subspace with strong correlation, less redundancy, and robust distinguishability, our OSR includes three stages, self-latent-aware, orthogonal subspace-aware, and structure representation-aware, respectively. First, the self-latent-aware stage promotes the latent subspace strongly correlated with the data space to discover interpretable factors, but with poor independence of variation. Second, the following orthogonal subspace-aware stage adaptively learns some 1-D linear subspace spanned by a set of orthogonal bases in the latent space. There is less redundancy between them, expressing the corresponding independence. Third, the structure representation-aware stage aligns the projection on the orthogonal subspace and the latent variables. Accordingly, feature representation in each linear subspace can be distinguishable, enhancing the independent expression of interpretable factors. In addition, we design an alternating optimization step, achieving a tradeoff training of OSRGAN on different properties. Despite it strictly constrains orthogonality, the loss weight coefficient of distinguishability induced by orthogonality could be adjusted and balanced with correlation constraint. To elucidate, this tradeoff training prevents our OSRGAN from overemphasizing any property and damaging the expressiveness of the feature representation. It takes into account both interpretable factors and their independent variation characteristics. Meanwhile, alternating optimization could keep the cost and efficiency of forward inference unchanged and will not burden the computational complexity. In theory, we clarify the significance of OSR, which brings better independence of factors, along with interpretability as correlation could converge to a high range faster. Moreover, through the convergence behavior analysis, including the objective functions under different constraints and the evaluation curve with iterations, our model demonstrates enhanced stability and definitely converges toward a higher peak for disentanglement. To depict the performance in downstream tasks, we compared the state-of-the-art GAN-based and even VAE-based approaches on different datasets. Our OSRGAN achieves higher disentanglement scores on FactorVAE, SAP, MIG, and VP metrics. All the experimental results illustrate that our novel GAN-based framework has considerable advantages on disentanglement.

Robust Supervised and Semisupervised Least Squares Regression Using ℓ2,p-Norm Minimization

Least squares regression (LSR) is widely applied in statistics theory due to its theoretical solution, which can be used in supervised, semisupervised, and multiclass learning. However, LSR begins to fail and its discriminative ability cannot be guaranteed when the original data have been corrupted and noised. In reality, the noises are unavoidable and could greatly affect the error construction in LSR. To cope with this problem, a robust supervised LSR (RSLSR) is proposed to eliminate the effect of noises and outliers. The loss function adopts l2,p -norm ( ) instead of square loss. In addition, the probability weight is added to each sample to determine whether the sample is a normal point or not. Its physical meaning is very clear, in which if the point is normal, the probability value is 1; otherwise, the weight is 0. To effectively solve the concave problem, an iterative algorithm is introduced, in which additional weights are added to penalize normal samples with large errors. We also extend RSLSR to robust semisupervised LSR (RSSLSR) to fully utilize the limited labeled samples. A large number of classification performances on corrupted data illustrate the robustness of the proposed methods.

Missing-view completion for fatty liver disease detection

Fatty liver disease is a common disease that causes extra fat storage in an individual's liver. Patients with fatty liver disease may progress to cirrhosis and liver failure, further leading to liver cancer. The prevalence of fatty liver disease ranges from 10% to 30% in many countries. In general, detecting fatty liver requires professional neuroimaging modalities or methods such as computed tomography, ultrasound, and medical experts' practical experiences. Considering this point, finding intelligent electronic noninvasive diagnostic approaches are desired at present. Currently, most existing works in the area of computerized noninvasive disease detection often apply one view (modality) or perform multi-view (several modalities) analysis, e.g., face, tongue, and/or sublingual for disease detection. The multi-view data of patients provides more complementary information for diagnosis. However, due to the conditions of data acquisition, interference by human factors, etc., many multi-view data are defective with some missing-view information, making these multi-view data difficult to evaluate. This factor largely affects the performance of classifying disease and the development of fully computerized noninvasive methods. Thus, the purpose of this study is to address the missing view issue among noninvasive disease detection. In this work, a multi-view dataset containing facial, sublingual vein, and tongue images are initially processed to produce corresponding feature for incomplete multi-view disease diagnostic evaluation. Hereby, we propose a novel method, i.e., multi-view completion, to process the incomplete multi-view data in order to complete the missing-view information for classifying fatty liver disease from healthy candidates. In particular, this method can explore the intra-view and inter-view information to produce the missing-view data effectively. Extensive experiments on a collected dataset with 220 fatty liver patients and 220 healthy samples show that our proposed approach achieves better diagnostic results with missing-view completion compared to the original incomplete multi-view data under various classifiers. Related results prove that our method can effectively process the missing-view issue and improve the noninvasive disease detection performance.

Fuzzy Sparse Deviation Regularized Robust Principal Component Analysis

Robust principal component analysis (RPCA) is a technique that aims to make principal component analysis (PCA) robust to noise samples. The current modeling approaches of RPCA were proposed by analyzing the prior distribution of the reconstruction error terms. However, these methods ignore the influence of samples with large reconstruction errors, as well as the valid information of these samples in principal component space, which will degrade the ability of PCA to extract the principal component of data. In order to solve this problem, Fuzzy sparse deviation regularized robust principal component Analysis (FSD-PCA) is proposed in this paper. First, FSD-PCA learns the principal components by minimizing the square of l₂ -norm-based reconstruction error. Then, FSD-PCA introduces sparse deviation on reconstruction error term to relax the samples with large bias, thus FSD-PCA can process noise and principal components of samples separately as well as improve the ability of FSD-PCA for retaining the principal component information. Finally, FSD-PCA estimates the prior probability of each sample by fuzzy weighting based on the relaxed reconstruction error, which can improve the robustness of the model. The experimental results indicate that the proposed model performs excellent robustness against different types of noise than the state-of-art algorithms, and the sparse deviation term enables FSD-PCA to process noise information and principal component information separately, so FSD-PCA can filter the noise information of an image and restore the corrupted image.

Relaxed Block-Diagonal Dictionary Pair Learning With Locality Constraint for Image Recognition

We propose a novel structured analysis-synthesis dictionary pair learning method for efficient representation and image classification, referred to as relaxed block-diagonal dictionary pair learning with a locality constraint (RBD-DPL). RBD-DPL aims to learn relaxed block-diagonal representations of the input data to enhance the discriminability of both analysis and synthesis dictionaries by dynamically optimizing the block-diagonal components of representation, while the off-block-diagonal counterparts are set to zero. In this way, the learned synthesis subdictionary is allowed to be more flexible in reconstructing the samples from the same class, and the analysis dictionary effectively transforms the original samples into a relaxed coefficient subspace, which is closely associated with the label information. Besides, we incorporate a locality-constraint term as a complement of the relaxation learning to enhance the locality of the analytical encoding so that the learned representation exhibits high intraclass similarity. A linear classifier is trained in the learned relaxed representation space for consistent classification. RBD-DPL is computationally efficient because it avoids both the use of class-specific complementary data matrices to learn discriminative analysis dictionary, as well as the time-consuming l₁/l₀ -norm sparse reconstruction process. The experimental results demonstrate that our RBD-DPL achieves at least comparable or better recognition performance than the state-of-the-art algorithms. Moreover, both the training and testing time are significantly reduced, which verifies the efficiency of our method. The MATLAB code of the proposed RBD-DPL is available at https://github.com/chenzhe207/RBD-DPL.

Robust Multi-Label Classification with Enhanced Global and Local Label Correlation

Data representation is of significant importance in minimizing multi-label ambiguity. While most researchers intensively investigate label correlation, the research on enhancing model robustness is preliminary. Low-quality data is one of the main reasons that model robustness degrades. Aiming at the cases with noisy features and missing labels, we develop a novel method called robust global and local label correlation (RGLC). In this model, subspace learning reconstructs intrinsic latent features immune from feature noise. The manifold learning ensures that outputs obtained by matrix factorization are similar in the low-rank latent label if the latent features are similar. We examine the co-occurrence of global and local label correlation with the constructed latent features and the latent labels. Extensive experiments demonstrate that the classification performance with integrated information is statistically superior over a collection of state-of-the-art approaches across numerous domains. Additionally, the proposed model shows promising performance on multi-label when noisy features and missing labels occur, demonstrating the robustness of multi-label classification.

A supervised discriminant data representation: application to pattern classification

The performance of machine learning and pattern recognition algorithms generally depends on data representation. That is why, much of the current effort in performing machine learning algorithms goes into the design of preprocessing frameworks and data transformations able to support effective machine learning. The method proposed in this work consists of a hybrid linear feature extraction scheme to be used in supervised multi-class classification problems. Inspired by two recent linear discriminant methods: robust sparse linear discriminant analysis (RSLDA) and inter-class sparsity-based discriminative least square regression (ICS_DLSR), we propose a unifying criterion that is able to retain the advantages of these two powerful methods. The resulting transformation relies on sparsity-promoting techniques both to select the features that most accurately represent the data and to preserve the row-sparsity consistency property of samples from the same class. The linear transformation and the orthogonal matrix are estimated using an iterative alternating minimization scheme based on steepest descent gradient method and different initialization schemes. The proposed framework is generic in the sense that it allows the combination and tuning of other linear discriminant embedding methods. According to the experiments conducted on several datasets including faces, objects, and digits, the proposed method was able to outperform competing methods in most cases.

Learning Salient and Discriminative Descriptor for Palmprint Feature Extraction and Identification

Palmprint recognition has been widely applied in security and, particularly, authentication. In the past decade, various palmprint recognition methods have been proposed and achieved promising recognition performance. However, most of these methods require rich a priori knowledge and cannot adapt well to different palmprint recognition scenarios, including contact-based, contactless, and multispectral palmprint recognition. This problem limits the application and popularization of palmprint recognition. In this article, motivated by the least square regression, we propose a salient and discriminative descriptor learning method (SDDLM) for general scenario palmprint recognition. Different from the conventional palmprint feature extraction methods, the SDDLM jointly learns noise and salient information from the pixels of palmprint images, simultaneously. The learned noise enforces the projection matrix to learn salient and discriminative features from each palmprint sample. Thus, the SDDLM can be adaptive to multiscenarios. Experiments were conducted on the IITD, CASIA, GPDS, PolyU near infrared (NIR), noisy IITD, and noisy GPDS palmprint databases, and palm vein and dorsal hand vein databases. It can be seen from the experimental results that the proposed SDDLM consistently outperformed the classical palmprint recognition methods and state-of-the-art methods for palmprint recognition.

Dual Shared-Specific Multiview Subspace Clustering

Multiview subspace clustering has received significant attention as the availability of diverse of multidomain and multiview real-world data has rapidly increased in the recent years. Boosting the performance of multiview clustering algorithms is challenged by two major factors. First, since original features from multiview data are highly redundant, reconstruction based on these attributes inevitably results in inferior performance. Second, since each view of such multiview data may contain unique knowledge as against the others, it remains a challenge to exploit complimentary information across multiple views while simultaneously investigating the uniqueness of each view. In this paper, we present a novel dual shared-specific multiview subspace clustering (DSS-MSC) approach that simultaneously learns the correlations between shared information across multiple views and also utilizes view-specific information to depict specific property for each independent view. Further, we formulate a dual learning framework to capture shared-specific information into the dimensional reduction and self-representation processes, which strengthens the ability of our approach to exploit shared information while preserving view-specific property effectively. The experimental results on several benchmark datasets have demonstrated the effectiveness of the proposed approach against other state-of-the-art techniques.

Linear embedding by joint Robust Discriminant Analysis and Inter-class Sparsity

Linear Discriminant Analysis (LDA) and its variants are widely used as feature extraction methods. They have been used for different classification tasks. However, these methods have some limitations that need to be overcome. The main limitation is that the projection obtained by LDA does not provide a good interpretability for the features. In this paper, we propose a novel supervised method used for multi-class classification that simultaneously performs feature selection and extraction. The targeted projection transformation focuses on the most discriminant original features, and at the same time, makes sure that the transformed features (extracted features) belonging to each class have common sparsity. Our proposed method is called Robust Discriminant Analysis with Feature Selection and Inter-class Sparsity (RDA_FSIS). The corresponding model integrates two types of sparsity. The first type is obtained by imposing the ℓ_2,1 constraint on the projection matrix in order to perform feature selection. The second type of sparsity is obtained by imposing the inter-class sparsity constraint used for ensuring a common sparsity structure in each class. An orthogonal matrix is also introduced in our model in order to guarantee that the extracted features can retain the main variance of the original data and thus improve the robustness to noise. The proposed method retrieves the LDA transformation by taking into account the two types of sparsity. Various experiments are conducted on several image datasets including faces, objects and digits. The projected features are used for multi-class classification. Obtained results show that the proposed method outperforms other competing methods by learning a more compact and discriminative transformation.

Sketch-Based Subspace Clustering of Hyperspectral Images

Sparse subspace clustering (SSC) techniques provide the state-of-the-art in clustering of hyperspectral images (HSIs). However, their computational complexity hinders their applicability to large-scale HSIs. In this paper, we propose a large-scale SSC-based method, which can effectively process large HSIs while also achieving improved clustering accuracy compared to the current SSC methods. We build our approach based on an emerging concept of sketched subspace clustering, which was to our knowledge not explored at all in hyperspectral imaging yet. Moreover, there are only scarce results on any large-scale SSC approaches for HSI. We show that a direct application of sketched SSC does not provide a satisfactory performance on HSIs but it does provide an excellent basis for an effective and elegant method that we build by extending this approach with a spatial prior and deriving the corresponding solver. In particular, a random matrix constructed by the Johnson-Lindenstrauss transform is first used to sketch the self-representation dictionary as a compact dictionary, which significantly reduces the number of sparse coefficients to be solved, thereby reducing the overall complexity. In order to alleviate the effect of noise and within-class spectral variations of HSIs, we employ a total variation constraint on the coefficient matrix, which accounts for the spatial dependencies among the neighbouring pixels. We derive an efficient solver for the resulting optimization problem, and we theoretically prove its convergence property under mild conditions. The experimental results on real HSIs show a notable improvement in comparison with the traditional SSC-based methods and the state-of-the-art methods for clustering of large-scale images.

Leveraging Coupled Interaction for Multimodal Alzheimer's Disease Diagnosis

As the population becomes older worldwide, accurate computer-aided diagnosis for Alzheimer's disease (AD) in the early stage has been regarded as a crucial step for neurodegeneration care in recent years. Since it extracts the low-level features from the neuroimaging data, previous methods regarded this computer-aided diagnosis as a classification problem that ignored latent featurewise relation. However, it is known that multiple brain regions in the human brain are anatomically and functionally interlinked according to the current neuroscience perspective. Thus, it is reasonable to assume that the extracted features from different brain regions are related to each other to some extent. Also, the complementary information between different neuroimaging modalities could benefit multimodal fusion. To this end, we consider leveraging the coupled interactions in the feature level and modality level for diagnosis in this paper. First, we propose capturing the feature-level coupled interaction using a coupled feature representation. Then, to model the modality-level coupled interaction, we present two novel methods: 1) the coupled boosting (CB) that models the correlation of pairwise coupled-diversity on both inconsistently and incorrectly classified samples between different modalities and 2) the coupled metric ensemble (CME) that learns an informative feature projection from different modalities by integrating the intrarelation and interrelation of training samples. We systematically evaluated our methods with the AD neuroimaging initiative data set. By comparison with the baseline learning-based methods and the state-of-the-art methods that are specially developed for AD/MCI (mild cognitive impairment) diagnosis, our methods achieved the best performance with accuracy of 95.0% and 80.7% (CB), 94.9% and 79.9% (CME) for AD/NC (normal control), and MCI/NC identification, respectively.

Learning Latent Low-Rank and Sparse Embedding for Robust Image Feature Extraction

To defy the curse of dimensionality, the inputs are always projected from the original high-dimensional space into the target low-dimension space for feature extraction. However, due to the existence of noise and outliers, the feature extraction task for corrupted data is still a challenging problem. Recently, a robust method called low rank embedding (LRE) was proposed. Despite the success of LRE in experimental studies, it also has many disadvantages: 1) The learned projection cannot quantitatively interpret the importance of features. 2) LRE does not perform data reconstruction so that the features may not be capable of holding the main energy of the original "clean" data. 3) LRE explicitly transforms error into the target space. 4) LRE is an unsupervised method, which is only suitable for unsupervised scenarios. To address these problems, in this paper, we propose a novel method to exploit the latent discriminative features. In particular, we first utilize an orthogonal matrix to hold the main energy of the original data. Next, we introduce an l_2,1 -norm term to encourage the features to be more compact, discriminative and interpretable. Then, we enforce a columnwise l_2,1 -norm constraint on an error component to resist noise. Finally, we integrate a classification loss term into the objective function to fit supervised scenarios. Our method performs better than several state-of-the-art methods in terms of effectiveness and robustness, as demonstrated on six publicly available datasets.

Discriminative Sparsity Graph Embedding for Unconstrained Face Recognition

In this paper, we propose a new dimensionality reduction method named Discriminative Sparsity Graph Embedding (DSGE) which considers the local structure information and the global distribution information simultaneously. Firstly, we adopt the intra-class compactness constraint to automatically construct the intrinsic adjacent graph, which enhances the reconstruction relationship between the given sample and the non-neighbor samples with the same class. Meanwhile, the inter-class compactness constraint is exploited to construct the penalty adjacent graph, which reduces the reconstruction influence between the given sample and the pseudo-neighbor samples with the different classes. Then, the global distribution constraints are introduced to the projection objective function for seeking the optimal subspace which compacts intra-classes samples and alienates inter-classes samples at the same time. Extensive experiments are carried out on AR, Extended Yale B, LFW and PubFig databases which are four representative face datasets, and the corresponding experimental results illustrate the effectiveness of our proposed method.

Low-Rank Preserving Projection Via Graph Regularized Reconstruction

Preserving global and local structures during projection learning is very important for feature extraction. Although various methods have been proposed for this goal, they commonly introduce an extra graph regularization term and the corresponding regularization parameter that needs to be tuned. However, tuning the parameter manually not only is time-consuming, but also is difficult to find the optimal value to obtain a satisfactory performance. This greatly limits their applications. Besides, projections learned by many methods do not have good interpretability and their performances are commonly sensitive to the value of the selected feature dimension. To solve the above problems, a novel method named low-rank preserving projection via graph regularized reconstruction (LRPP_GRR) is proposed. In particular, LRPP_GRR imposes the graph constraint on the reconstruction error of data instead of introducing the extra regularization term to capture the local structure of data, which can greatly reduce the complexity of the model. Meanwhile, a low-rank reconstruction term is exploited to preserve the global structure of data. To improve the interpretability of the learned projection, a sparse term with l_2,1 norm is imposed on the projection. Furthermore, we introduce an orthogonal reconstruction constraint to make the learned projection hold main energy of data, which enables LRPP_GRR to be more flexible in the selection of feature dimension. Extensive experimental results show the proposed method can obtain competitive performance with other state-of-the-art methods.

Approximate Low-Rank Projection Learning for Feature Extraction

Feature extraction plays a significant role in pattern recognition. Recently, many representation-based feature extraction methods have been proposed and achieved successes in many applications. As an excellent unsupervised feature extraction method, latent low-rank representation (LatLRR) has shown its power in extracting salient features. However, LatLRR has the following three disadvantages: 1) the dimension of features obtained using LatLRR cannot be reduced, which is not preferred in feature extraction; 2) two low-rank matrices are separately learned so that the overall optimality may not be guaranteed; and 3) LatLRR is an unsupervised method, which by far has not been extended to the supervised scenario. To this end, in this paper, we first propose to use two different matrices to approximate the low-rank projection in LatLRR so that the dimension of obtained features can be reduced, which is more flexible than original LatLRR. Then, we treat the two low-rank matrices in LatLRR as a whole in the process of learning. In this way, they can be boosted mutually so that the obtained projection can extract more discriminative features. Finally, we extend LatLRR to the supervised scenario by integrating feature extraction with the ridge regression. Thus, the process of feature extraction is closely related to the classification so that the extracted features are discriminative. Extensive experiments are conducted on different databases for unsupervised and supervised feature extraction, and very encouraging results are achieved in comparison with many state-of-the-arts methods.