Discriminative Dictionary Learning Research Articles

Hierarchical Discriminative Deep Dictionary Learning

In deep dictionary learning multiple dictionaries are learned based on information at various levels of abstraction. We propose a novel hierarchical discriminative dictionary learning layer embedded within a neural network with an image classification objective. Discrimination is induced in the learned synthesis dictionaries at multiple hierarchical levels in a simple way by a one-hot-code representation of the class labels during the training backward pass. In addition, local sparse representation objectives are approximated during the forward pass, introducing local regularization. We evaluate our proposal on five known datasets and we either outperform state-of-the-art methods or achieve competitive classification results.

ELM embedded discriminative dictionary learning for image classification

Dictionary learning is a widely adopted approach for image classification. Existing methods focus either on finding a dictionary that produces discriminative sparse representation, or on enforcing priors that best describe the dataset distribution. In many cases, the dataset size is often small with large intra-class variability and nondiscriminative feature space. In this work we propose a simple and effective framework called ELM-DDL to address these issues. Specifically, we represent input features with Extreme Learning Machine (ELM) with orthogonal output projection, which enables diverse representation on nonlinear hidden space and task specific feature learning on output space. The embeddings are further regularized via a maximum margin criterion (MMC) to maximize the inter-class variance and minimize intra-class variance. For dictionary learning, we design a novel weighted class specific ℓ1,2 norm to regularize the sparse coding vectors, which promotes uniformity of the sparse patterns of samples belonging to the same class and suppresses support overlaps of different classes. We show that such regularization is robust, discriminative and easy to optimize. The proposed method is combined with a sparse representation classifier (SRC) to evaluate on benchmark datasets. Results show that our approach achieves state-of-the-art performance compared to other dictionary learning methods.

A Survey of Dictionary Learning in Medical Image Analysis and Its Application for Glaucoma Diagnosis

Dictionary learning has shown its effectiveness in computer vision with the concise expression form but the powerful representation. Dictionary learning represents images with a bag of visual words (BoVW), which is a collection of atoms expressively representative for images. Recently, several task-specific dictionary learning methods have been proposed and successfully applied in medical image analysis, such as de-noising, classification, segmentation, and so on, which promotes the development of computer-aided diagnosis. In this paper, first we give a survey for dictionary learning-based medical image analysis methods including: (1) three discriminative dictionary learning frameworks, (2) CT image de-noising based on dictionary learning, and (3) histopathological image classification using sparse representation. Then, a novel method named Low-rank Shared Dictionary Learning (LRSDL), is presented to achieve accurate glaucoma diagnosis on fundus images. The LRSDL generates a shared codebook for image reconstruction and a particular one to handle the difference between the healthy and glaucomatous images. Benefit from this strategy, LRSDL not only possess distinct glaucoma-related features, but also share common patterns among all the fundus images. Experimental results show that the method effectively delivers glaucoma diagnosis with the accuracy of 92.90%. This endows dictionary learning method a great potential for glaucoma diagnosis and proves the feasibility of its application to medical image analysis.

Person Re-identification by Cross-View Discriminative Dictionary Learning with Metric Embedding

Person Re-identification by Cross-View Discriminative Dictionary Learning with Metric Embedding

COMP-07. MACHINE LEARNING PREDICTS PROGRESSION AND SURVIVAL IN GLIOMA USING RADIOHISTOGENOMIC FEATURES

Abstract BACKGROUND Random forest classifiers can predict survival and progression status with radiomic features. Here, we compare the predictions of radiomic, histomic, and genomic features on progression-free survival (PFS) and overall survival (OS) in patients with low-grade glioma (LGG). METHODS We identified patients from our institution and The Cancer Genome Atlas (TCGA) with LGG and analyzed histopathology, mRNA expression, and MRI sequences. Differential gene expression analysis was performed to obtain top gene expression values. Discriminative feature-oriented dictionary learning (DFDL)1 was used to extract histomic features from histopathology slides. Forty-four computer-derived texture features were extracted from segmented tumors for each MRI sequence. A random forest classifier with cross-validation was then used to evaluate binary predictions of PFS and OS using radiomic, genomic, and histomic features from institutional and TCGA data. RESULTS The out-of-bag (OOB) error rates for PFS and OS predictions from combined institutional and TCGA radiomic features (n = 104) was 0.22 and 0.15, respectively. Similarly, OOB error rates of PFS and OS predictions for institutional histomic features (n = 66) was 0.24 and 0.29, respectively. OOB error rates for predictions of PFS and OS from combined institutional and TCGA genomic features (n = 283) were 0.14 and 0.24, respectively. A combined model incorporating radiogenomic features from institutional and TCGA data performed slightly worse than either modality alone (PFS OOB error = 0.28, OS OOB error = 0.21), possibly due to fewer cases (n = 67). CONCLUSION Our findings indicate combining radiomic and genomic features can predict PFS and OS in patients with LGG. The utility of this methodology may help to predict malignant transformation in LGG. REFERENCES: Vu TH, Mousavi HS, Monga V, Rao G, Rao UA. Histopathological image classification using discriminative feature-oriented dictionary learning. IEEE transactions on medical imaging. 2015;35(3):738–51.

Discriminative Eigenpixels-Based Dictionary Learning for Hyperspectral Image Classification

Sparse representation (SR) model has been applied to hyperspectral image (HSI) classification based on the observation that any spectral pixel could be approximately represented by a linear combination of several common pixels, but its discriminative ability is not deeply explored due to an insufficient description on spatial–spectral information and less emphasis on the dictionary structure. In this letter, we propose a new algorithm of HSI classification based on discriminative eigenpixels-based dictionary learning. Instead of using neighbor pixels directly, we define a new homogeneous region for each pixel, respectively, to exploit more spatial–spectral information and extract eigenpixels from homogeneous regions to preserve the essentials for each class. For the representation-based model, a discriminative eigenpixels-based dictionary is learned in homogeneous regions, where the locality of pixels is exploited to enhance the discriminative ability. Finally, we code a homogeneous region associated with a query pixel on the learned dictionary to determine its label by use of both the nearest neighbor (NN) classifier and majority voting. Experiments are conducted on four HSIs to demonstrate the effectiveness of the proposed method.

Brain status modeling with non-negative projective dictionary learning

Accurate prediction of individuals’ brain age is critical to establish a baseline for normal brain development. This study proposes to model brain development with a novel non-negative projective dictionary learning (NPDL) approach, which learns a discriminative representation of multi-modal neuroimaging data for predicting brain age. Our approach encodes the variability of subjects in different age groups using separate dictionaries, projecting features into a low-dimensional manifold such that information is preserved only for the corresponding age group. The proposed framework improves upon previous discriminative dictionary learning methods by incorporating orthogonality and non-negativity constraints, which remove representation redundancy and perform implicit feature selection. We study brain development on multi-modal brain imaging data from the PING dataset (N = 841, age = 3−21 years). The proposed analysis uses our NDPL framework to predict the age of subjects based on cortical measures from T1-weighted MRI and connectome from diffusion weighted imaging (DWI). We also investigate the association between age prediction and cognition, and study the influence of gender on prediction accuracy. Experimental results demonstrate the usefulness of NDPL for modeling brain development.

Structure-constrained discriminative dictionary learning based on Schatten p-norm for face recognition

Dictionary learning based on low-rank representations has been widely studied in recent years and has achieved impressive performance in face recognition. However, studies on the case where both the training and testing data are polluted with large noise are still in their infancy. In this paper, to capture the essential structure hidden in noisy data and to learn a discriminative dictionary, we propose a novel model of structure-constrained discriminative dictionary learning based on the Schatten p-norm (SDDLSp), in which (1) the nonconvex Schatten p-norm (0<p<1) is used to approximate the rank minimization problem, (2) a weighted representation regularization term is constructed to generate a coefficient matrix with a class-wise block-diagonal structure, and (3) a linear classifier based on the coefficient is simultaneously optimized with the dictionary and representation update. Thus, the learned dictionary and representation perform well in face recognition problems, particularly with training and testing images that are both polluted by large noises. An iterative algorithm to solve the proposed SDDLSp is also presented in this paper. Experiments on four face databases demonstrate the advantages of the proposed method over previous dictionary-learning methods.

Low-rank graph preserving discriminative dictionary learning for image recognition

Discriminative dictionary learning plays a key role in sparse representation-based classification. In this paper, we propose a low-rank graph preserving discriminative dictionary learning (LRGPDDL) method for sparse representation-based image recognition. Specifically, we learn a common sub-dictionary as well as several class-specific sub-dictionaries to explicitly capture the common information shared by all the classes and the class-specific information belonging to the corresponding class. We also impose a low-rank constraint on each sub-dictionary to weaken the negative influence from noise contained in training samples. A discriminative graph preserving criterion and a discriminative reconstruction error term are used for exploiting discriminative information, which can improve the discriminative ability of the learned dictionary effectively. In addition, an incoherence term is also introduced into the proposed dictionary learning model to encourage the learned sub-dictionaries to be as independent as possible. Experimental results on several image datasets verify the effectiveness and robustness of LRGPDDL.

Robust Distracter-Resistive Tracker via Learning a Multi-Component Discriminative Dictionary

Discriminative dictionary learning (DDL) provides an appealing paradigm for appearance modeling in visual tracking. However, most existing DDL-based trackers cannot handle drastic appearance changes, especially for scenarios with background cluster and/or similar object interference. One reason is that they often suffer from the loss of subtle visual information, which is critical to distinguish an object from distracters. In this paper, we explore the use of activations from the convolutional layer of a convolutional neural network to improve the object representation and then propose a robust distracter-resistive tracker via learning a multi-component discriminative dictionary. The proposed method exploits both the intra-class and inter-class visual information to learn shared atoms and the class-specific atoms. By imposing several constraints into the objective function, the learned dictionary is reconstructive, compressive, and discriminative, and thus can better distinguish an object from the background. In addition, our convolutional features have structural information for object localization and balance the discriminative power and semantic information of the object. Tracking is carried out within a Bayesian inference framework where a joint decision measure is used to construct the observation model. To alleviate the drift problem, the reliable tracking results obtained online are accumulated to update the dictionary. Both the qualitative and quantitative results on the CVPR2013 benchmark, the VOT2015 data set, and the SPOT data set demonstrate that our tracker achieves substantially better overall performance against the state-of-the-art approaches.

SAR Automatic Target Recognition Based on Attribute Scattering Center Model and Discriminative Dictionary Learning

Synthetic aperture radar (SAR) automatic target recognition plays a crucial role in both national defense and civil applications. Although many methods have been proposed, it remains one of the most challenging tasks to identify different terrain vehicle targets. In this paper, a novel method is proposed based on the attribute scattering center model (ASCM) and discriminative dictionary learning. This method contains three main stages. In the first stage, the low-level local features are generated by the convolution operation between SAR patches and the ASCMs with different parameters. These parameters are selected by genetic algorithm in advance with labeled patches. In the feature coding stage, a discriminative dictionary learning method is proposed with the label consistent and locality constraint on the sparse coefficient to incorporate the label information and local geometric information into the sparse representation, which is called label consistent and locality constraint discriminative dictionary learning (LcLcDDL) method. The low-level local features are sparsely coded using the dictionary learned by LcLcDDL method. In the feature pooling stage, the spatial pyramid matching method with max pooling is utilized to aggregate the coding coefficients of the low-level local features to generate the high-level global feature for each patch. The generated high-level global feature is feed into the linear support vector machine to recognize which classes the patch belongs to. Experimental results conducted on the moving and stationary target acquisition and recognition data set show that the performance of the proposed method is better than several available approaches.

Discriminative dictionary learning for retinal vessel segmentation using fusion of multiple features

In recent years, automated retinal vessel segmentation has become especially essential for the early detection of some ophthalmological and cardiovascular diseases. In this paper, we have presented a new retinal vessel segmentation method via discriminative dictionary learning using fusion of multiple features, which is able to capture both thick and thin vessel structures. In the training stage, we employ six different enhancement algorithms to obtain multiple complementary features that contain rich vascular information. Then, the manually annotated ground-truth vessels are classified into thick or thin vessels as the label information, and the label consistent KSVD based framework is applied to train the dictionary for vessel segmentation. In the testing stage, comprehensive experiments are conducted on three datasets to measure segmentation performance with eight representative evaluation metrics. The average sensitivity reaches 0.7915, 0.7560 and 0.7202 respectively, suggesting that our method can segment tiny vascular structures well.

Myoelectric Signal Classification of Targeted Muscles Using Dictionary Learning.

Surface electromyography (sEMG) signals comprise electrophysiological information related to muscle activity. As this signal is easy to record, it is utilized to control several myoelectric prostheses devices. Several studies have been conducted to process sEMG signals more efficiently. However, research on optimal algorithms and electrode placements for the processing of sEMG signals is still inconclusive. In addition, very few studies have focused on minimizing the number of electrodes. In this study, we investigated the most effective method for myoelectric signal classification with a small number of electrodes. A total of 23 subjects participated in the study, and the sEMG data of 14 different hand movements of the subjects were acquired from targeted muscles and untargeted muscles. Furthermore, the study compared the classification accuracy of the sEMG data using discriminative feature-oriented dictionary learning (DFDL) and other conventional classifiers. DFDL demonstrated the highest classification accuracy among the classifiers, and its higher quality performance became more apparent as the number of channels decreased. The targeted method was superior to the untargeted method, particularly when classifying sEMG signals with DFDL. Therefore, it was concluded that the combination of the targeted method and the DFDL algorithm could classify myoelectric signals more effectively with a minimal number of channels.

Dictionary learning enhancement framework: Learning a non-linear mapping model to enhance discriminative dictionary learning methods

In this paper, a new framework is presented to enhance the reconstruction and discrimination capabilities of existing discriminative dictionary learning methods. In the proposed framework, a non-linear mapping model is introduced to learn a feature space in a way that any standard discriminative dictionary learning algorithms could achieve higher classification accuracies. The proposed feature mapping process targets to boost the standard dictionary learning methods by facilitating their optimization process. The mapping model uses a modified autoencoder network to provide a higher level of reconstruction and discrimination capabilities for the discriminative dictionary learning methods. The proposed dictionary learning enhancement (DLE) framework could be applied to any discriminative dictionary learning methods with the embedded discriminative term in their objective functions. Our experiments on several real-world image datasets demonstrate that the proposed framework could improve the classification accuracies of standard discriminative dictionary learning methods.

Discriminative Fisher Embedding Dictionary Learning Algorithm for Object Recognition.

Both interclass variances and intraclass similarities are crucial for improving the classification performance of discriminative dictionary learning (DDL) algorithms. However, existing DDL methods often ignore the combination between the interclass and intraclass properties of dictionary atoms and coding coefficients. To address this problem, in this paper, we propose a discriminative Fisher embedding dictionary learning (DFEDL) algorithm that simultaneously establishes Fisher embedding models on learned atoms and coefficients. Specifically, we first construct a discriminative Fisher atom embedding model by exploring the Fisher criterion of the atoms, which encourages the atoms of the same class to reconstruct the corresponding training samples as much as possible. At the same time, a discriminative Fisher coefficient embedding model is formulated by imposing the Fisher criterion on the profiles (row vectors of the coding coefficient matrix) and coding coefficients, which forces the coding coefficient matrix to become a block-diagonal matrix. Since the profiles can indicate which training samples are represented by the corresponding atoms, the proposed two discriminative Fisher embedding models can alternatively and interactively promote the discriminative capabilities of the learned dictionary and coding coefficients. The extensive experimental results demonstrate that the proposed DFEDL algorithm achieves superior performance in comparison with some state-of-the-art dictionary learning algorithms on both hand-crafted and deep learning-based features.

Discriminative dictionary learning for local LV wall motion classification in cardiac MRI

The characterization of cardiac function is of high clinical interest for early diagnosis and better patient follow-up in cardiovascular diseases. A large number of cardiac image analysis methods and more precisely in cine-Magnetic Resonance Imaging (MRI) have been proposed to quantify both shape and motion parameters. However, the first major problem to address lies in the cardiac image segmentation that is most often needed to extract the myocardium before any other process such as motion tracking, or registration. Moreover, intelligent systems based on classification and learning techniques have emerged over the last years in medical imaging. In this paper, a new method is proposed to help medical experts in classifying the Left Ventricle (LV) wall motion without the need of image segmentation and through the learning of motion features by using dictionary learning techniques. Specifically the novelty of this approach lies in the extraction of new spatio-temporal descriptors and in the use of discriminative dictionary learning (DL) techniques to classify normal/abnormal LV function in cardiac MRI. Local radial spatio-temporal profiles are first extracted from bidimensional (2D) short axis (SAX) MRI images, for each anatomical segment of the LV cavity. These profiles inherently contain discriminatory information that can help for cardiac motion characterization. An advantage of this approach is that these profiles are constructed from a very limited user interaction that corresponds to a number of five points in only one frame of the sequence, (without the need of LV boundaries segmentation) and by exploiting all the phases of the cardiac cycle. Two specific discriminative DL algorithms have been selected for the LV wall classification based on these profiles: Label Consistent K-SVD (LC-KSVD) and Fisher Discriminative (FD-DL). For the application of the proposed methods, cine-MR SAX images have been collected from a control group of 9 healthy subjects and from 9 patients with cardiac dyssynchrony. Radial strain curves in 2D Speckle Tracking Echocardiography (2D-STE) have been analysed for the patient group and have been used as reference truth. They allowed to label each profile as normal or abnormal. The best performance has been achieved in the Wavelet domain by the LC-KSVD algorithm with an accuracy of 84.07% in the classification of radial spatio-temporal profiles and using a leave-one-out patient cross validation. The approach has been compared with recent methods of the literature and offers a good compromise between performance, user interaction, time computing and complexity. This new method of LV classification, with minimal user interaction and based on discriminative DL has not been previously reported. It could help to improve the performance of pre-screening systems for cardiac assessment, which can affect positively the quality of the early diagnosis for heart failure patients.

Ship Detection for PolSAR Images via Task-Driven Discriminative Dictionary Learning

Ship detection with polarimetric synthetic aperture radar (PolSAR) has received increasing attention for its wide usage in maritime applications. However, extracting discriminative features to implement ship detection is still a challenging problem. In this paper, we propose a novel ship detection method for PolSAR images via task-driven discriminative dictionary learning (TDDDL). An assumption that ship and clutter information are sparsely coded under two separate dictionaries is made. Contextual information is considered by imposing superpixel-level joint sparsity constraints. In order to amplify the discrimination of the ship and clutter, we impose incoherence constraints between the two sub-dictionaries in the objective of feature coding. The discriminative dictionary is trained jointly with a linear classifier in task-driven dictionary learning (TDDL) framework. Based on the learnt dictionary and classifier, we extract discriminative features by sparse coding, and obtain robust detection results through binary classification. Different from previous methods, our ship detection cue is obtained through active learning strategies rather than artificially designed rules, and thus, is more adaptive, effective and robust. Experiments performed on synthetic images and two RADARSAT-2 images demonstrate that our method outperforms other comparative methods. In addition, the proposed method yields better shape-preserving ability and lower computation cost.

Multi-layer discriminative dictionary learning with locality constraint for image classification

Discriminative dictionary learning (DDL) has demonstrated significantly improved performance for image classification. However, most of the existing DDL methods just adopt the single-layer dictionary learning architecture, which narrows the discriminative ability of the coding vectors. Another limitation of these methods is that the atoms of the learned dictionary are easily affected by the noise in the original data. To this end, a powerful architecture, called the multi-layer discriminative dictionary learning (MDDL) with locality constraint, is proposed for image classification. Through the multi-layer dictionary learning, the robust dictionary is obtained in the final layer, where the separability of coding vectors from different classes is also increased. Meanwhile, benefiting from joint classifier training and multi-layer dictionary learning, the discriminability of the learned coding vectors is further enhanced. Besides, by utilizing the graph Laplacian matrices based on the learned dictionaries, not only the locality information of the original data is preserved, but also it can avoid very large values in the coding vectors to reduce the test error caused by overfitting. In addition, an iterative algorithm is devised to efficiently solve the proposed MDDL. The experimental results demonstrate that our methods can achieve promising classification results on well-known benchmark image datasets.

Coherence-regularized discriminative dictionary learning for histopathological image classification

In this paper, a novel coherence-regularized discriminative dictionary learning (CRDDL) algorithm is proposed to deal with histopathological image classification. By incorporating two constructed special regularization terms into an objective function, i.e., the self-coherence within each intra-class dictionary and the mutual coherence between inter-class dictionaries, high-quality discriminative healthy and diseased dictionaries can both be explicitly learned. Furthermore, to balance the reconstruction and discrimination abilities of learned dictionaries, we minimize the reconstruction error of intra-class samples and maximize the reconstruction error of inter-class samples. Finally, reconstruction error vectors are employed to design the classifier of histopathological images. Experimental results demonstrate the improved performance of the proposed CRDDL algorithm in comparison with other previously reported discriminative dictionary learning algorithms.

Jointly discriminative projection and dictionary learning for domain adaptive collaborative representation-based classification

In recent years, collaborative representation-based classification (CRC) methods have shown impressive performance in many recognition tasks. However, when the training data have different distributions with the testing data, the performance of CRC will be degraded significantly. On the other hand, concatenating training data from different sources as a single data set will affect the performance of CRC, as the shift exists between the different source domains. To address these problems, in this paper, we propose a Jointly Discriminative projection and Dictionary learning for domain adaptive Collaborative Representation-based Classification method (JD2-CRC). As the distributions of different source domains may be dissimilar, the data from all domains are projected into a common feature subspace where the latent shared structures can be found. Then a compact dictionary is learned to represent the projected data well. To find the most suitable projection matrices and dictionary for CRC, we design the objective function of JD2-CRC,according to the classification rule of CRC in feature subspace, which minimizes the ratio of within-class reconstruction errors over between-class reconstruction errors. Different to traditional optimization methods, an effective optimization procedure is presented based on gradient descent. Thus, the obtained collaborative representations have a better discriminability and suit the classification rule of CRC well. The experimental results demonstrate that the proposed method can achieve superior performance against other state-of-the-art methods.