Local Feature Selection Research Articles

Learning Local Quality-Aware Structures of Salient Regions for Stereoscopic Images via Deep Neural Networks

The perceptual quality of stereoscopic images plays an essential role in the human perception of visual information. However, most available stereoscopic image quality assessment (SIQA) methods evaluate 3D visual experience using hand-crafted features or shallow architectures, which cannot model the visual properties of stereo images well. In this paper, we use convolutional neural networks (CNNs) to learn deeper local quality-aware structures for stereo images. With different inputs, two CNN models are designed for no-reference SIQA tasks. The one-column CNN model directly accepts a cyclopean view as the input, and the three-column CNN model jointly considers the cyclopean, left and right views as CNN inputs. The two SIQA frameworks share the same implementation approach: First, to overcome the obstacle of limited SIQA datasets, we accept image patches that have been cropped from corresponding stereopairs as inputs for local quality-sensitive feature extraction. Next, a local feature selection algorithm is used to remove related features on non-salient patches, which could cause large prediction errors. Finally, the reserved local visual structures of salient regions are aggregated into a final quality score in an end-to-end manner. Experimental results on three public SIQA databases demonstrate that our method outperforms most state-of-the-art no-reference (NR) SIQA methods. The results of a cross-database experiment also show the robustness and generality of the proposed method.

Gaussian discriminative component analysis for early detection of Alzheimer’s disease: A supervised dimensionality reduction algorithm

BackgroundUsing multiple modalities of biomarkers, several machine leaning-based approaches have been proposed to characterize patterns of structural, functional and metabolic differences discernible from multimodal neuroimaging data for Alzheimer’s disease (AD). Current investigations report several studies using binary classification often augmented with local feature selection methods, while fewer other studies address the challenging problem of multiclass classification. New methodTo assess the merits of each of these research directions, this study introduces a supervised Gaussian discriminative component analysis (GDCA) algorithm, which can effectively delineate subtle changes of early mild cognitive impairment (EMCI) group in relation to the cognitively normal control (CN) group. Using 251 CN, 297 EMCI, 196 late MCI (LMCI), and 162 AD subjects from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) and considering both structural and functional (metabolic) information from magnetic resonance imaging (MRI) and positron emission tomography (PET) modalities as input, the proposed method conducts a dimensionality reduction algorithm taking into consideration the interclass information to define an optimal eigenspace that maximizes the discriminability of selected eigenvectors. ResultsThe proposed algorithm achieves an accuracy of 79.25 % for delineating EMCI from CN using 38.97 % of Gaussian discriminative components (i.e., dimensionality reduction). Moreover, for detecting the different stages of AD, a multiclass classification experiment attained an overall accuracy of 67.69 %, and more notably, discriminates MCI and AD groups from the CN group with an accuracy of 75.28 % using 48.90 % of the Gaussian discriminative components. Comparison with existing method(s)The classification results of the proposed GDCA method outperform the more recently published state-of-the-art methods in AD-related multiclass classification tasks, and seems to be the most stable and reliable in terms of relating the most relevant features to the optimal classification performance. ConclusionThe proposed GDCA model with its high prospects for multiclass classification has a high potential for deployment as a computer aided clinical diagnosis system for AD.

The effects of globalisation techniques on feature selection for text classification

Text classification (TC) is very important and critical task in the 21th century as there exist high volume of electronic data on the Internet. In TC, textual data are characterised by a huge number of highly sparse features/terms. A typical TC consists of many steps and one of the most important steps is undoubtedly feature selection (FS). In this study, we have comprehensively investigated the effects of various globalisation techniques on local feature selection (LFS) methods using datasets with different characteristics such as multi-class unbalanced (MCU), multi-class balanced (MCB), binary-class unbalanced (BCU) and binary-class balanced (BCB). The globalisation techniques used in this study are summation (SUM), weighted-sum (AVG), and maximum (MAX). To investigate the effect of globalisation techniques, we used three LFS methods named as Discriminative Feature Selection (DFSS), odds ratio (OR) and chi-square (CHI2). In the experiments, we have utilised four different benchmark datasets named as Reuters-21578, 20Newsgroup., Enron1, and Polarity in addition to Support Vector Machines (SVM) and Decision Tree (DT) classifiers. According to the experimental results, the most successful globalisation technique is AVG while all situations are taken into account. The experimental results indicate that DFSS method is more successful than OR and CHI2 methods on datasets with MCU and MCB characteristics. However, CHI2 method seems more accurate than OR and DFSS methods on datasets with BCU and BCB characteristics. Also, SVM classifier performed better than DT classifier in most cases.

LFSpy: A Python Implementation of Local Feature Selection for Data Classification with scikit-learn Compatibility

LFSpy: A Python Implementation of Local Feature Selection for Data Classification with scikit-learn Compatibility

A Local Neighborhood Robust Fuzzy Clustering Image Segmentation Algorithm Based on an Adaptive Feature Selection Gaussian Mixture Model.

Since the fuzzy local information C-means (FLICM) segmentation algorithm cannot take into account the impact of different features on clustering segmentation results, a local fuzzy clustering segmentation algorithm based on a feature selection Gaussian mixture model was proposed. First, the constraints of the membership degree on the spatial distance were added to the local information function. Second, the feature saliency was introduced into the objective function. By using the Lagrange multiplier method, the optimal expression of the objective function was solved. Neighborhood weighting information was added to the iteration expression of the classification membership degree to obtain a local feature selection based on feature selection. Each of the improved FLICM algorithm, the fuzzy C-means with spatial constraints (FCM_S) algorithm, and the original FLICM algorithm were then used to cluster and segment the interference images of Gaussian noise, salt-and-pepper noise, multiplicative noise, and mixed noise. The performances of the peak signal-to-noise ratio and error rate of the segmentation results were compared with each other. At the same time, the iteration time and number of iterations used to converge the objective function of the algorithm were compared. In summary, the improved algorithm significantly improved the ability of image noise suppression under strong noise interference, improved the efficiency of operation, facilitated remote sensing image capture under strong noise interference, and promoted the development of a robust anti-noise fuzzy clustering algorithm.

Adaptive neural tree exploiting expert nodes to classify high-dimensional data

Classification of high dimensional data suffers from curse of dimensionality and over-fitting. Neural tree is a powerful method which combines a local feature selection and recursive partitioning to solve these problems, but it leads to high depth trees in classifying high dimensional data. On the other hand, if less depth trees are used, the classification accuracy decreases or over-fitting increases. This paper introduces a novel Neural Tree exploiting Expert Nodes (NTEN) to classify high-dimensional data. It is based on a decision tree structure, whose internal nodes are expert nodes performing multi-dimensional splitting. Any expert node has three decision-making abilities. Firstly, they can select the most eligible neural network with respect to the data complexity. Secondly, they evaluate the over-fitting. Thirdly, they can cluster the features to jointly minimize redundancy and overlapping. To this aim, metaheuristic optimization algorithms including GA, NSGA-II, PSO and ACO are applied. Based on these concepts, any expert node splits a class when the over-fitting is low, and clusters the features when the over-fitting is high. Some theoretical results on NTEN are derived, and experiments on 35 standard data show that NTEN reaches good classification results, reduces tree depth without over-fitting and degrading accuracy.

Local feature selection based on artificial immune system for classification

Conventional feature selection algorithms select a global feature subset for the entire sample space. In contrast, in this paper we propose an efficient filter local feature selection algorithm based on artificial immune system, which assigns a locally relevant feature subset for each neighboring region of the sample space. This algorithm introduces a clonal selection algorithm to explore the search space for the optimal feature subsets, and adopts local clustering idea as an evaluation criterion that maximizes the inter-class distance and minimizes the intra-class distance in the small region of each sample. Experimental results on a wide variety of synthetic and UCI datasets demonstrates that our proposed method achieves better performance than both state-of-the-art global feature selection algorithms and local feature selection algorithms. In addition, a main parameter analysis of the proposed method is carried out.

Deep Regionlets: Blended Representation and Deep Learning for Generic Object Detection.

In this article, we propose a novel object detection algorithm named "Deep Regionlets" by integrating deep neural networks and a conventional detection schema for accurate generic object detection. Motivated by the effectiveness of regionlets for modeling object deformations and multiple aspect ratios, we incorporate regionlets into an end-to-end trainable deep learning framework. The deep regionlets framework consists of a region selection network and a deep regionlet learning module. Specifically, given a detection bounding box proposal, the region selection network provides guidance on where to select sub-regions from which features can be learned from. An object proposal typically contains three - 16 sub-regions. The regionlet learning module focuses on local feature selection and transformations to alleviate the effects of appearance variations. To this end, we first realize non-rectangular region selection within the detection framework to accommodate variations in object appearance. Moreover, we design a "gating network" within the regionlet leaning module to enable instance dependent soft feature selection and pooling. The Deep Regionlets framework is trained end-to-end without additional efforts. We present ablation studies and extensive experiments on the PASCAL VOC dataset and the Microsoft COCO dataset. The proposed method yields competitive performance over state-of-the-art algorithms, such as RetinaNet and Mask R-CNN, even without additional segmentation labels.

Stepwise detection and evaluation reveal miR-10b and miR-222 as a remarkable prognostic pair for glioblastoma

Despite the existence of many clinical and molecular factors reported that contribute to survival in glioblastoma, prevailing studies fell into partial or local feature selection for survival analysis. We proposed a feature selection strategy including not only joint covariate detection but also its evaluations, and performed it on miRNA expression profiles with glioblastoma. MiR-10b and miR-222 were selected as the most significant two-dimensional feature. Crucially, we integrated in vitro experiments on GBM cells and in vivo studies on a mouse model of human glioma to elucidate the synergistic effects between miR-10b and miR-222. Inhibition of miR-10b and miR-222 strongly suppress GBM cells growth, invasion, and induce apoptosis by co-targeting PTEN and leading to activation of p53 ultimately. We also demonstrated that miR-10b and miR-222 co-target BIM to induce apoptosis independent of p53 status. The results define mir-10b and mir-222 important roles in gliomagenesis and provided a reliable survival analysis strategy.

An efficient approach for dimension selection and classification in HSI images

ABSTRACTIn this paper, we have proposed a divide and select approach for dimension selection (DSCS) and classification of hyperspectral images (HSI). The DSCS algorithm has been offered using a two-stage framework. In stage-1, we have generated the intermediate channels for every pair of spectral dimensions that are weakly correlated and highly significant. These channels have derived by splitting the adjacent dimensions without any loss of physical meaning. Intermediated channels are concatenated with original HSI in a better sense to transform the image into a more informative datacube. However, this leads to an increase in the dimensionality of the image. In stage-2, a trace and determinant based local feature response approach has applied to select the most informative dimensions of transformed HSI. We have exploited the local feature and scale selection methods to obtain significant channels. Finally, classification experiments have conducted for selected bands with SVM (Support Vector Machine) and LDA (Linear Discriminant Analysis). An expansion graph cut optimization has applied to improve the classification accuracy. This method has demonstrated as state-of-art for target detection in hyperspectral scenes.

Exploring influencing factors on transit ridership from a local perspective

Purpose Exploring the influencing factors on urban rail transit (URT) ridership is vital for travel demand estimation and urban resources planning. Among various existing ridership modeling methods, direct demand model with ordinary least square (OLS) multiple regression as a representative has considerable advantages over the traditional four-step model. Nevertheless, OLS multiple regression neglects spatial instability and spatial heterogeneity from the magnitude of the coefficients across the urban area. This paper aims to focus on modeling and analyzing the factors influencing metro ridership at the station level. Design/methodology/approach This paper constructs two novel direct demand models based on geographically weighted regression (GWR) for modeling influencing factors on metro ridership from a local perspective. One is GWR with globally implemented LASSO for feature selection, and the other one is geographically weighted LASSO (GWL) model, which is GWR with locally implemented LASSO for feature selection. Findings The results of real-world case study of Shenzhen Metro show that the two local models presented perform better than the traditional global model (OLS) in terms of estimation error of ridership and goodness-of-fit. Additionally, the GWL model results in a better fit than GWR with global LASSO model, indicating that the locally implemented LASSO is more effective for the accurate estimation of Shenzhen metro ridership than global LASSO does. Moreover, the information provided by both two local models regarding the spatial varied elasticities demonstrates the strong spatial interpretability of models and potentials in transport planning. Originality/value The main contributions are threefold: the approach is based on spatial models considering spatial autocorrelation of variables, which outperform the traditional global regression model – OLS – in terms of model fitting and spatial explanatory power. GWR with global feature selection using LASSO and GWL is compared through a real-world case study on Shenzhen Metro, that is, the difference between global feature selection and local feature selection is discussed. Network structures as a type of factors are quantified with the measurements in the field of complex network.

Modeling of oil near-infrared spectroscopy based on similarity and transfer learning algorithm

Near-infrared spectroscopy mainly reflects the frequency-doubled and total-frequency absorption information of hydrogen-containing groups (O‒H, C‒H, N‒H, S‒H) in organic molecules for near-infrared lights with different wavelengths, so it is applicable to testing of most raw materials and products in the field of petrochemicals. However, the modeling process needs to collect a large number of laboratory analysis data. There are many oil sources in China, and oil properties change frequently. Modeling of each raw material is not only unfeasible but also will affect its engineering application efficiency. In order to achieve rapid modeling of near-infrared spectroscopy and based on historical data of different crude oils under different detection conditions, this paper discusses about the feasibility of the application of transfer learning algorithm and makes it possible that transfer learning can assist in rapid modeling using certain historical data under similar distributions under a small quantity of new data. In consideration of the requirement of transfer learning for certain similarity of different datasets, a transfer learning method based on local similarity feature selection is proposed. The simulation verification of spectral data of 13 crude oils measured by three different probe detection methods is performed. The effectiveness and application scope of the transfer modeling method under different similarity conditions are analyzed.

A fast algorithm for local feature selection in data classification

AbstractTypical feature selection methods select a global feature subset that is applied over all regions of the sample space. In localized feature selection (LFS), each region of the sample space is associated with its own optimized feature subset. This allows the feature subset to adapt to local variations in the sample space. Feature subsets are selected such that within a localized region, within‐class distances are minimized and between‐class distances are maximized. LFS outperforms global feature selection methods. LFS is solved using a randomized rounding approach when weights of regions are fixed. Randomized rounding is a too time‐consuming algorithm. In this paper, we show that LFS has a closed‐form solution when weights of regions are fixed. Using this closed‐form solution can decrease the runtime of solving LFS substantially. Experimental results on real datasets confirm that the classification error rate of our proposed method and the randomized rounding‐based method are the same; the runtime of our proposed method is much better than that of the randomized rounding‐based method; and the classification error rate of our proposed method and the randomized rounding‐based method outperforms the state‐of‐the‐art feature selection methods.

TOWARDS AN AUTOMATIC DIAGNOSIS SYSTEM FOR LUMBAR DISC HERNIATION: THE SIGNIFICANCE OF LOCAL SUBSET FEATURE SELECTION

Herniation in the lumbar area is one of the most common diseases which results in lower back pain (LBP) causing discomfort and inconvenience in the patients’ daily lives. A computer aided diagnosis (CAD) system can be of immense benefit as it generates diagnostic results within a short time while increasing precision of diagnosis and eliminating human errors. We have proposed a new method for automatic diagnosis of lumbar disc herniation based on clinical MRI data. We use T2-W sagittal and myelograph images. The presented method has been applied on 30 clinical cases, each containing 7 discs (210 lumbar discs) for the herniation diagnosis. We employ Otsu thresholding method to extract the spinal cord from MR images of lumbar disc. A third order polynomial is then aligned on the extracted spinal cords, and by the end of preprocessing stage, all the T2-W sagittal images will have been prepared for specifying disc boundary and labeling. Having extracted an ROI for each disc, we proceed to use intensity and shape features for classification. The extracted features have been selected by Local Subset Feature Selection. The results demonstrated 91.90%, 92.38% and 95.23% accuracy for artificial neural network, K-nearest neighbor and support vector machine (SVM) classifiers respectively, indicating the superiority of the proposed method to those mentioned in similar studies.

Spam Review Classification Using Ensemble of Global and Local Feature Selectors

Abstract In our work, we propose an ensemble of local and global filter-based feature selection method to reduce the high dimensionality of feature space and increase accuracy of spam review classification. These selected features are then used for training various classifiers for spam detection. Experimental results with four classifiers on two available datasets of hotel reviews show that the proposed feature selector improves the performance of spam classification in terms of well-known performance metrics such as AUC score.

A novel approach to emotion recognition using local subset feature selection and modified Dempster-Shafer theory

BackgroundEmotion recognition is an increasingly important field of research in brain computer interactions.IntroductionWith the advance of technology, automatic emotion recognition systems no longer seem far-fetched. Be that as it may, detecting neural correlates of emotion has remained a substantial bottleneck. Settling this issue will be a breakthrough of significance in the literature.MethodsThe current study aims to identify the correlations between different emotions and brain regions with the help of suitable electrodes. Initially, independent component analysis algorithm is employed to remove artifacts and extract the independent components. The informative channels are then selected based on the thresholded average activity value for obtained components. Afterwards, effective features are extracted from selected channels common between all emotion classes. Features are reduced using the local subset feature selection method and then fed to a new classification model using modified Dempster-Shafer theory of evidence.ResultsThe presented method is employed to DEAP dataset and the results are compared to those of previous studies, which highlights the significant ability of this method to recognize emotions through electroencephalography, by the accuracy of about 91%. Finally, the obtained results are discussed and new aspects are introduced.ConclusionsThe present study addresses the long-standing challenge of finding neural correlates between human emotions and the activated brain regions. Also, we managed to solve uncertainty problem in emotion classification which is one of the most challenging issues in this field. The proposed method could be employed in other practical applications in future.

Medical Image Fusion using bi-dimensional empirical mode decomposition (BEMD) and an Efficient Fusion Scheme.

Background: Medical image fusion is being widely used for capturing complimentary information from images of different modalities. Combination of useful information presented in medical images is the aim of image fusion techniques, and the fused image will exhibit more information in comparison with source images.Objective: In the current study, a BEMD-based multi-modal medical image fusion technique is utilized. Moreover, Teager-Kaiser energy operator (TKEO) was applied to lower BIMFs. The results were compared to six routine methods.Material and Methods: In this study, which is of experimental type, an image fusion technique using bi-dimensional empirical mode decomposition (BEMD), Teager-Kaiser energy operator (TKEO) as a local feature selection and Hierarchical Model And X (HMAX) model is presented. BEMD fusion technique can preserve much functional information. In the process of fusion, we adopt the fusion rule of TKEO for lower bi-dimensional intrinsic mode functions (BIMFs) of two images and HMAX visual cortex model as a fusion rule for higher BIMFs, which are verified to be more appropriate for human vision system. Integrating BEMD and this efficient fusion scheme can retain more spatial and functional features of input images.Results: We compared our method with IHS, DWT, LWT, PCA, NSCT and SIST methods. The simulation results and fusion performance show that the presented method is effective in terms of mutual information, quality of fused image (QAB/F), standard deviation, peak signal to noise ratio, structural similarity and considerably better results compared to six typical fusion methods.Conclusion: The statistical analyses revealed that our algorithm significantly improved spatial features and diminished the color distortion compared to other fusion techniques. The proposed approach can be used for routine practice. Fusion of functional and morphological medical images is possible before, during and after treatment of tumors in different organs. Image fusion can enable interventional events and can be further assessed.

Student’s t-Hidden Markov Model for Unsupervised Learning Using Localized Feature Selection

Recently, the hidden Markov model (HMM) with student’s t-mixture model (SMM), called student’s t-HMM (SHMM) for short, has received much attention in unsupervised learning of sequential data. However, the current existing SHMMs fail to take into consideration of the relevant features embedded in local subspaces, thus influencing their performances in clustering. To address the problem, a novel SHMM is proposed by combining the measure of localized feature saliency (LFS) with SMM and utilizing two student’s t-distributions as subcomponents to respectively describe the distributions of useful features and non-salient “features,” with the purpose of accurately modeling the hidden state observation emission distributions of SHMM. Moreover, we exploit the variational Bayesian learning technique to simultaneously estimate the LFS, the number of components and other parameters of the herein proposed SHMM. Experimental results on both synthetic and real data sets demonstrate the improved robustness, effectiveness, and accuracy of our model.

Prediction of paroxysmal Atrial Fibrillation: A machine learning based approach using combined feature vector and mixture of expert classification on HRV signal

Background and ObjectiveParoxysmal Atrial Fibrillation (PAF) is one of the most common major cardiac arrhythmia. Unless treated timely, PAF might transform into permanent Atrial Fibrillation leading to a high rate of morbidity and mortality. Therefore, increasing attention has been directed towards prediction of PAF, to enable early detection and prevent further progression of the disease. Notwithstanding the pharmacological and electrical treatments, a validated method to predict the onset of PAF is yet to be developed. We aim to address this issue through integrating classical and modern methods. MethodsTo increase the predictivity, we have made use of a combination of features extracted through linear, time-frequency, and nonlinear analyses performed on heart rate variability. We then apply a novel approach to local feature selection using meticulous methodologies, developed in our previous works, to reduce the dimensionality of the feature space. Subsequently, the Mixture of Experts classification is employed to ensure a precise decision-making on the output of different processes. In the current study, we analyzed 106 signals from 53 pairs of ECG recordings obtained from the standard database called Atrial Fibrillation Prediction Database (AFPDB). Each pair of data contains one 30-min ECG segment that ends just before the onset of PAF event and another 30-min ECG segment at least 45 min distant from the onset. ResultsCombining the features that are extracted using both classical and modern analyses was found to be significantly more effective in predicting the onset of PAF, compared to using either analyses independently. Also, the Mixture of Experts classification yielded more precise class discrimination than other well-known classifiers. The performance of the proposed method was evaluated using the Atrial Fibrillation Prediction Database (AFPDB) which led to sensitivity, specificity, and accuracy of 100%, 95.55%, and 98.21% respectively. ConclusionPrediction of PAF has been a matter of clinical and theoretical importance. We demonstrated that utilising an optimized combination of — as opposed to being restricted to — linear, time-frequency, and nonlinear features, along with applying the Mixture of Experts, contribute greatly to an early detection of PAF, thus, the proposed method is shown to be superior to those mentioned in similar studies in the literature.

Opto‐electric target tracking algorithm based on local feature selection and particle filter optimization

SummaryAiming at exploring the opto‐electric target tracking, which is an important technology in the field of computer vision, the binocular stereo vision camera opto‐electric target tracking is studied and and a multi feature fusion characterization modeling method locally weighted is proposed. The target area is divided into multiple sub‐image areas by the modeling method, the feature histogram after the background weighting is extracted, and the sub‐image region is taken as a basic unit for adjusting the feature weight. The sub‐image area selected is regarded as the significant area, and the significant area is further extracted and fused in particle filter tracking algorithm. Then, the obtained significant are is conducted with color distribution processing. In the state prediction stage, the Mean Shift algorithm is applied to optimize each particle so that it converges to the optimal position. The experiment results showed that the multi feature fusion representation modeling method has better tracking accuracy and stability compared with the traditional fusion method and after the color distribution treatment; it has strong anti‐jamming for background effect. It is concluded that using Mean Shift algorithm for particle optimization can further strengthen the accurate tracking of the targets.