Search

It is hard to construct an optimal classifier for high-dimensional imbalanced data, on which the performance of classifiers is seriously affected and becomes poor. Although many approaches, such as resampling, cost-sensitive, and ensemble learning methods, have been proposed to deal with the skewed data, they are constrained by high-dimensional data with noise and redundancy. In this study, we propose an adaptive subspace optimization ensemble method (ASOEM) for high-dimensional imbalanced data classification to overcome the above limitations. To construct accurate and diverse base classifiers, a novel adaptive subspace optimization (ASO) method based on adaptive subspace generation (ASG) process and rotated subspace optimization (RSO) process is designed to generate multiple robust and discriminative subspaces. Then a resampling scheme is applied on the optimized subspace to build a class-balanced data for each base classifier. To verify the effectiveness, our ASOEM is implemented based on different resampling strategies on 24 real-world high-dimensional imbalanced datasets. Experimental results demonstrate that our proposed methods outperform other mainstream imbalance learning approaches and classifier ensemble methods.

Recently, multilabel classification has generated considerable research interest. However, the high dimensionality of multilabel data incurs high costs; moreover, in many real applications, a number of labels of training samples are randomly missed. Thus, multilabel classification can have great complexity and ambiguity, which means some feature selection methods exhibit poor robustness and yield low prediction accuracy. To solve these issues, this paper presents a novel feature selection method based on multilabel fuzzy neighborhood rough sets (MFNRS) and maximum relevance minimum redundancy (MRMR) that can be used on multilabel data with missing labels. First, to handle multilabel data with missing labels, a relation coefficient of samples, label complement matrix, and label-specific feature matrix are constructed and implemented in a linear regression model to recover missing labels. Second, the margin-based fuzzy neighborhood radius, fuzzy neighborhood similarity relationship, and fuzzy neighborhood information granule are developed. The MFNRS model is built based on multilabel neighborhood rough sets combined with fuzzy neighborhood rough sets. Based on algebra and information views, certain fuzzy neighborhood entropy-based uncertainty measures are proposed for MFNRS. The fuzzy neighborhood mutual information-based MRMR model with label correlation is improved to evaluate the performance of candidate features. Finally, a feature selection algorithm is designed to improve the performance for multilabel data with missing labels. Experiments on twenty datasets verify that our method is effective not only for recovering missing labels but also for selecting significant features with better classification performance.

Shannon’s entropy or an extension of Shannon’s entropy can be used to quantify information transmission between or among variables. Mutual information is the pair-wise information that captures nonlinear relationships between variables. It is more robust than linear correlation methods. Beyond mutual information, two generalizations are defined for multivariate distributions: interaction information or co-information and total correlation or multi-mutual information. In comparison to mutual information, interaction information and total correlation are underutilized and poorly studied in applied neuroscience research. Quantifying information flow between brain regions is not explicitly explained in neuroscience by interaction information and total correlation. This article aims to clarify the distinctions between the neuroscience concepts of mutual information, interaction information, and total correlation. Additionally, we proposed a novel method for determining the interaction information between three variables using total correlation and conditional mutual information. On the other hand, how to apply it properly in practical situations. We supplied both simulation experiments and real neural studies to estimate functional connectivity in the brain with the above three higher-order information-theoretic approaches. In order to capture redundancy information for multivariate variables, we discovered that interaction information and total correlation were both robust, and it could be able to capture both well-known and yet-to-be-discovered functional brain connections.

Vacating room after encryption (VRAE) is a popular framework of reversible data hiding for encrypted images (RDHEI). Most VRAE based RDHEI methods do not make a desirable payload. To address this issue, this paper proposes a novel data hiding technique using adaptive difference recovery (ADR) and exploits this novel technique to design an efficient VRAE based RDHEI method. Specifically, the proposed ADR-based data hiding technique adaptively determines the range of original difference between a cover pixel and a reference pixel with correlation according to the marked difference generated by bit substitution technique. The cover pixel can be recovered losslessly with the range of original difference and the reference pixel. In the proposed RDHEI method, the encrypted image is generated by block permutation and block-based modulation so that spatial redundancy within encrypted image blocks is preserved. Then the ADR-based data hiding technique is applied to the encrypted blocks. Experiment results show that the proposed RDHEI method outperforms some state-of-the-art RDHEI methods in payload. The average embedding rates of the proposed RDHEI method on the BOSSbase, BOWS-2 and UCID datasets are 3.2045, 3.1145 and 2.4633 bpp, respectively.

Heart attack else wise termed as myocardial infarction (MI) causes irreparable death of cardiac muscles yielding the focal reason for most casualties among all cardiovascular diseases (CVDs’). A 12-lead electrocardiogram (ECG) generally depicts cardiac abnormalities and so customary deep learning (DL) approaches use the whole signal for binary detection purposes, that is separating healthy control (HC), and MI classes. This paper proposes an alternative approach where 21 temporal features in lieu of the temporal signal are collected from the 12 lead data to reduce redundancy and class imbalance keeping the vital information intact. Then these extracted features are fed into a detection model consisting of a one dimensional (1-D) convolutional neural network (CNN) and a bidirectional long short-term memory (bi-LSTM) layer which classifies into three classes, namely: HC, MI, and non-myocardial infarction (non-MI) subjects for a realistic and reliable assessment. The model’s performance is evaluated using 517 records acquired from the Physikalisch-Technische Bundesanstalt (PTB) database and a state-of-art overall accuracy of 99.246%, kappa of 0.983, and macro averaged F1 score of 98.86% were achieved using stratified 5-fold cross-validation. DL methods suffer to make unbiased decisions in the case of class imbalance due to an insufficient amount of data for a particular class and thus temporal features are employed to inherently reduce this problem. The successful performance of the extracted features depends on the precise detection of fiducial points, and so multiple novel algorithms have been introduced in this paper.

Feature selection is an important preprocessing process in machine learning. It selects the crucial features by removing irrelevant features or redundant features from the original feature set. Most of feature selection algorithms focus on maximizing relevant information and minimizing redundant information. In order to remove more redundant information in the evaluation criteria, we propose a feature selection based on mutual information with correlation coefficient (CCMI) in this paper. We introduce the correlation coefficient in the paper, and combine the correlation coefficient and mutual information to measure the relationship between different features. We use the absolute value of the correlation coefficient between two different features as the weight of the redundant item denoted by the mutual information in the evaluation standard. In order to select low redundancy features effectively, we also use the principle of minimization in the evaluation criteria. By comparing with 7 popular contrast algorithms in 12 data sets, CCMI has achieved the highest average classification accuracy for two classifiers of SVM and KNN. Experimental results show that our proposed CCMI has better feature classification capability.

One of the important tasks in the operating room is monitoring the depth of anesthesia (DoA) during surgery, and noninvasive techniques are very popular. Hence, we propose a new scheme for DoA monitoring considering the time-frequency analysis of electroencephalography (EEG) signals and GLCM features extracted from them. To this end, at first, the time-frequency map (TFM) of each channel of each EEG is computed by smoothed pseudo-Wigner–Ville distribution (SPWVD), where the EEG signal used in this paper is recorded in 15 channels. After that, we consider the gray-level co-occurrence matrix (GLCM) to obtain the content of TFM, and after that, four features such as homogeneity, correlation, energy, and contrast are obtained for each GLCM. Finally, after the selection of efficient features using the minimum redundancy maximum relevance (MRMR) method, the K-nearest neighbor (KNN) classifier is utilized to determine the DoA. Here, we consider the three states, namely, deep hypnotic, surgical anesthesia, and sedation and awake states according to bispectral index (BIS), and each EEG epoch is classified to these states. We also employ data augmentation to enhance the training phase and increase accuracy. We obtain the accuracy and confusion matrix of the proposed method. We also analyze the effects of a number of gray levels of GLCM, distance measure in KNN classifier, and parameters of data augmentation on the performance of the proposed method. Results indicate the efficiency of the proposed method to determine the DoA during surgery.

Feature selection based on entropy structure can be roughly divided into two types according to whether they are related to kernel density estimation (KDE). The first type is feature selection based on non-KDE entropy. The second type is feature selection based on KDE entropy. Compared with the first type, the second type avoids discretization and obtains more accurate mutual information when handling continuous data. However, existing feature selection methods based on KDE entropy neglect the fact that samples with noise have negative impacts on KDE. Besides, the feature evaluation functions don’t effectively assess relevance and redundancy of features. Thus, a feature selection method via maximizing independent and minimizing redundant classification information ratio is constructed based on adaptive weighted kernel density estimation. Firstly, an adaptive weighted kernel density estimation model is designed. Secondly, the entropy structure is defined by the adaptive weighted kernel density estimation model, and their theoretical properties are explored. Thirdly, a feature selection algorithm via maximizing independent and minimizing redundant classification information ratio is designed from the viewpoint of adaptive weighted kernel density estimation. Finally, comprehensive experiments are performed. The results illustrate that our approach has certain robustness, validity and superiority compared with other representative feature selection approaches.

To reduce the redundancy among different color channels, e.g., YUV, previous methods usually adopt a linear model that tends to be oversimple for complex image content. We propose a neural-network-based method for cross-channel prediction in intra frame coding. The proposed network utilizes twofold cues, i.e., the neighboring reconstructed samples with all channels, and the co-located reconstructed samples with partial channels. Specifically, for YUV video coding, the neighboring samples with YUV are processed by several fully connected layers; the co-located samples with Y are processed by convolutional layers; and the proposed network fuses the twofold cues. We observe that the integration of twofold information is crucial to the performance of intra prediction of the chroma components. We have designed the network architecture to achieve a good balance between compression performance and computational efficiency. Moreover, we propose a transform domain loss for the training of the network. The transform domain loss helps obtain more compact representations of residues in the transform domain, leading to higher compression efficiency. The proposed method is plugged into HEVC and VVC test models to evaluate its effectiveness. Experimental results show that our method provides more accurate cross-channel intra prediction compared with previous methods. On top of HEVC, our method achieves on average 1.3%, 5.4%, and 3.8% BD-rate reductions for Y, Cb, and Cr on common test sequences, and on average 3.8%, 11.3%, and 9.0% BD-rate reductions for Y, Cb, and Cr on ultra-high-definition test sequences. On top of VVC, our method achieves on average 0.5%, 1.7%, and 1.3% BD-rate reductions for Y, Cb, and Cr on common test sequences.

The Reversible data hiding (RDH) approach can retrieve the original image from the marked image without any distortion. RDH in encrypted images is an approach that hides extra information into the ciphertext using a skill of recovering the actual data losslessly. To guarantee reversibility for addressing the information redundancy drawback, the cover image pixels are copied into two images. This paper presents a high capacity RDH scheme in encrypted images using fuzzy-based encryption. Initially, the texture classification is processed by a convolutional neural network (CNN) to classify the dense and transparent region. It automatically identifies the significant features without any individual supervision. Then, the plain text encryption is activated by the fuzzy group teaching with infinite elliptic curve (FGTIE) method. To overcome the demerit of FCM, the GTA is hybrid with FCM approach and the encryption is processed by the IE method. Next, a new embedding approach is used to enhance the embedding capacity, namely quotient multi-pixel value differencing (QMPVD). In order to obtain the higher PSNR and payload, the multi-pixel differencing is hybrid with the quotient value differencing. Finally, the original data is extracted and recovered with good quality and high capacity. The performances are evaluated using several performance metrics such as PSNR, SSIM, BER, MSE, embedding capacity/payload, sensitivity, specificity, tampering ratio, correlation coefficient, number of pixel change rate and unified average changing intensity. The performance of PSNR and capacity is compared with existing approaches named Encrypted image-based RDH with Paillier cryptosystem (EIRDH-PC), EIRDH with Redundancy Transfer (EIRDH-RT) and EIRDH with pixel value ordering (EIRDH-PVO). The performance is calculated for three groups of images such as the brain, lungs and abdomen. The implementation results show that the introduced model attained better performance compared to existing approaches in terms of PSNR and capacity. Besides, the proposed approach achieved the merits of no pixel expansion, lossless and alternative order recovery.

This paper details the examination of a particular case of data compression, where the compression algorithm removes the redundancy from data, which occurs when edge-based compression algorithms compress (previously compressed) pixelated images. The newly created redundancy can be removed using another round of compression. This work utilized the JPEG-LS as an example of an edge-based compression algorithm for compressing pixelated images. The output of this process was subjected to another round of compression using a more robust but slower compressor (PAQ8f). The compression ratio of the second compression was, on average, 18%, which is high for random data. The results of the second compression were superior to the lossy JPEG. Under the used data set, lossy JPEG needs to sacrifice 10% on average to realize nearly total lossless compression ratios of the two-successive compressions. To generalize the results, fast general-purpose compression algorithms (7z, bz2, and Gzip) were used too.

Hyperspectral images can provide dozens to hundreds of continuous spectral bands, so the richness of information has been greatly improved. However, these bands lead to increasing complexity of data processing, and the redundancy of adjacent bands is large. Recently, although many band selection methods have been proposed, this task is rarely handled through the context information of the whole spectral bands. Moreover, the scholars mainly focus on the different numbers of selected bands to explain the influence by accuracy measures, neglecting how many bands to choose is appropriate. To tackle these issues, we propose a fast neighborhood grouping method for hyperspectral band selection (FNGBS). The hyperspectral image cube in space is partitioned into several groups using coarse-fine strategy. By doing so, it effectively mines the context information in a large spectrum range. Compared with most algorithms, the proposed method can obtain the most relevant and informative bands simultaneously as subset in accordance with two factors, such as local density and information entropy. In addition, our method can also automatically determine the minimum number of recommended bands by determinantal point process. Extensive experimental results on benchmark data sets demonstrate the proposed FNGBS achieves satisfactory performance against state-of-the-art algorithms.

Using a closed-form inverse kinematics solution for motion planning has many advantages compared to traditional numerical approaches, most notably much faster computation times and better suitability for real-time applications. Steady progress has been made to develop an analytic inverse kinematics solution for seven degrees of freedom (DoF) manipulators without joint offsets. Redundancy resolutions have been proposed to push the individual joint angles away from their limits. However, no further objectives are considered in these closed-form solutions. We propose a multi-objective real-time optimization approach that is able to minimize joint velocities and accelerations, while still avoiding joint limits. A closed-form solution to the optimization is derived, which only requires a single intuitive tuning parameter. We provide a free and open-source software implementation of the full kinematics algorithm that covers the complete solution space. The null-space motion resulting from our method is evaluated, achieving faster motion times and smoother arm motions along Cartesian paths in comparison to the state-of-the-art approach. Finally, experimental results with a KUKA iiwa robot are reported.

We study the consideration of fairness in redundant assignment for multi-agent task allocation. It has recently been shown that redundant assignment of agents to tasks provides robustness to uncertainty in task performance. However, the question of how to fairly assign these redundant resources across tasks remains unaddressed. In this paper, we present a novel problem formulation for fair redundant task allocation, which we cast as the optimization of worst-case task costs under a cardinality constraint. Solving this problem optimally is NP-hard. We exploit properties of supermodularity to propose a polynomial-time, near-optimal solution. In supermodular redundant assignment, the use of additional agents always improves task costs. Therefore, we provide a solution set that is $\alpha$ times larger than the cardinality constraint. This constraint relaxation enables our approach to achieve a super-optimal cost by using a sub-optimal assignment size. We derive the sub-optimality bound on this cardinality relaxation, $\alpha$. Additionally, we demonstrate that our algorithm performs near-optimally without the cardinality relaxation. We show simulations of redundant assignments of robots to goal nodes on transport networks with uncertain travel times. Empirically, our algorithm outperforms benchmarks, scales to large problems, and provides improvements in both fairness and average utility., Comment: (c) 2021 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works

A large composition of image, sound, motion, video and other signal sequences is always a rich source of useful and reliable information. With the proliferating cost and complexities of mass storage media, the need for redundancy and complex transmission-free signals is on the rise. Furthermore, the growing increase in the number of video-specific applications has heightened the need for digital multimedia technologies that emphasized bit compression. This paper presents a Discrete Transform Technique (DTT) that leveraged the integration of Discrete Cosine Transform (DCT) and Discrete Wavelet Transform (DWT) for higher scale video compression. The experimental study of the new technique was based on some online videos with varying sizes, sources and formats. The analysis of the experimental values obtained for standard performance metrics such as maximum difference, mean square error and peak signal to noise ratio for the pre- and post-compression videos portrayed a very significant and encouraging impact of the new technique. Comparative analysis based on compression ratios derived using the new technique and some recent video compression techniques also established a very impressive performance of the new technique.

Redundancy is an extremely significant challenge in data integration and decision making based on the model of soft set which is able to process data under uncertainty. There are four available methods which are designed to reduce the redundant parameters of the soft set. But there is a very low success rate on a large number of data sets obtained from practices by these methods. In order to overcome the inherent weakness, we propose a parameter reduction method based on chi square distribution for the model of soft set. Experimental results on two real-life application cases and thirty randomly generated data sets demonstrate that our algorithm largely improves the success rate of parameter reduction, redundant degree of parameter, and has higher practicability in comparison with the four existing normal parameter reduction algorithms.

We propose a neural network-based feature selection (FS) scheme that can control the level of redundancy in the selected features by integrating two penalties into a single objective function. The Group Lasso penalty aims to produce sparsity in features in a grouped manner. The redundancy-control penalty, which is defined based on a measure of dependence among features, is utilized to control the level of redundancy among the selected features. Both the penalty terms involve the $L_{2,1}$ -norm of weight matrix between the input and hidden layers. These penalty terms are nonsmooth at the origin, and hence, one simple but efficient smoothing technique is employed to overcome this issue. The monotonicity and convergence of the proposed algorithm are specified and proved under suitable assumptions. Then, extensive experiments are conducted on both artificial and real data sets. Empirical results explicitly demonstrate the ability of the proposed FS scheme and its effectiveness in controlling redundancy. The empirical simulations are observed to be consistent with the theoretical results.

In this letter, we propose soft compression, an lossless compression approach to shape coding for images using location index and codebook of designed shapes with various sizes. This method is different from traditional image compression methods, as it aims at finding the basic shape blocks of pictures to improve the compression ratio from the perspective of information theory and frequency. In particular, the triplets including location and codeword are stored when encoding an image instead of the entire array. It is shown that soft compression can greatly reduce the bandwidth and storage space needed in the process of transmitting and storing the same kind of images.

Light field (LF) imaging has generated considerable interest owing to its ability to capture both spatial and angular information of light rays simultaneously. However, the extremely large volume of data associated with LF imaging poses challenges to both data storage and transmission. This study addresses this issue by proposing a view synthesis-based LF image compression method using a generative adversarial network (GAN). The primary basis of compression relies on the fact that adjacent sub-aperture images (SAIs) are highly correlated. Accordingly, only sparsely sampled SAIs are transmitted and the others are reconstructed at the decoder side. The proposed sparse SAI sampling method enhances the quality of reconstructed SAIs by considering a fair trade-off between the number of SAIs available for use as priors in the synthesis process and SAI redundancy. The quality of reconstructed SAIs is further enhanced by a GAN-based SAI synthesis method, where the synthesis procedure is broken into disparity estimation and un-sampled SAI estimation components, and the adversarial nature of the jointly trained generative and discriminative networks results in a more accurate generative model. Furthermore, more texture details can be preserved in the synthesized SAIs by adopting a loss function in the GAN model based on perceptual quality. Extensive experimental results demonstrate the superiority of the proposed method relative to several other state-of-the-art compression methods in terms of standard quality metrics and the perceptual quality of the synthetic SAIs at the decoder side.

Recently, a reversible data hiding scheme by redundant space transfer has been proposed by Liu et al. In this scheme, the encryption process consists of bit plane disordering, sub-block position disordering and different-time Arnold transform in each sub-block. Then, the additional data is embedded into the LSB planes of the encrypted image by compressing the redundant space. However, to transfer the redundancy of the plain image into encrypted image, this scheme essentially just performs three types of disorder operations, which is vulnerable to security analysis. In this paper, we firstly demonstrate that the encryption method of this scheme can be broken by chosen-plaintext attack (CPA). Furthermore, in order to achieve the resistance against CPA, we introduce 2-D Logistic-adjusted-sine map (2D-LASM) to produce the encryption sequences with the control parameters generated from the plain image. Apart from that, we also improve the sparse matrix compression to achieve a superior embedding capacity. Extensive experiments and analyses demonstrate that the proposed scheme outperforms the original one in terms of both security and embedding capacity.

Advanced image scanning techniques produce high resolution medical images such as CT, MRI which in turn needs large storage space and bandwidth for transmitting over a network. Lossless compression is preferred for medical images to preserve important diagnostic details. However, it is only sufficient to maintain the high quality of an image in a diagnostically important region, namely Region of interest (ROI) for an accurate diagnosis. Non -ROI portion when compressed near-losslessly does not affect the image quality but reduces the file size effectively. We propose a compression technique where the prediction is done by Resolution Independent Gradient Edge Detector (RIGED) to de-correlate the image pixels and block-based arithmetic coding is used for encoding. The optimal threshold value, optimal q-level and the block-based coding removes inter-pixel, psycho-visual and coding redundancy from non-ROI part to achieve high compression whereas ROI part is compressed losslessly by removing inter-pixel and coding redundancy only. In this paper, optimal threshold-based predictive lossless compression in the ROI and optimal quantization (q) based near-lossless compression in the rest of the region is proposed. The proposed method is evaluated on volumetric 8 bit and 16 bit standard MR image data-set and validated on real patient’s 16 bit depth MR images collected from local hospitals. The performance of the proposed technique showed improvement over the existing techniques JPEG 2000, JPEG-LS, M-CALIC, JP3D, and CALIC by 40.89%, 34.50%, 32.92%, 22.36%, and 17.25% respectively in terms of Bits per Pixel (BPP).

Emission spectra yielded by laser-induced breakdown spectroscopy (LIBS) exhibit high dimensionality, redundancy, and sparsity. The high dimensionality is often addressed by principal component analysis (PCA) which creates a low dimensional embedding of the spectra by projecting them into the score space. However, PCA does not effectively deal with the sparsity of the analysed data, including LIBS spectra. Consequently, sparse PCA (SPCA) was proposed for the analysis of high-dimensional sparse data. Nevertheless, SPCA remains underutilized for LIBS applications. Thus, in this work, we show that SPCA combined with genetic algorithms offers marginal improvements in clustering and quantification using multivariate calibration. More importantly, we show that SPCA significantly improves the interpretability of loading spectra. In addition, we show that the loading spectra yielded by SPCA differ from those yielded by sparse partial least squares regression. Finally, by using the randomized SPCA (RSPCA) algorithm for carrying out SPCA, we indirectly demonstrate that the analysis of LIBS data can greatly benefit from the tools developed by randomized linear algebra: RSPCA offers a 20-fold increase in computation speed compared to PCA based on singular value decomposition.

Symbolic Regression searches for a mathematical expression that fits the input data set by minimizing the approximation error. The search space explored by this technique is composed of any mathematical function representable as an expression tree. This provides more flexibility for fitting the data but it also makes the task more challenging. The search space induced by this representation becomes filled with redundancy and ruggedness, sometimes requiring a higher computational budget in order to achieve good results. Recently, a new representation for Symbolic Regression was proposed, called Interaction-Transformation, which can represent function forms as a composition of interactions between predictors and the application of a single transformation function. In this work, we show how this representation can be modeled as a multi-layer neural network with the weights adjusted following the Extreme Learning Machine procedure. The results show that this approach is capable of finding equally good or better results than the current state-of-the-art with a smaller computational cost.

Hyperspectral target detection is widely used in both military and civilian fields. In practical applications, how to select a low-correlation and representative band subset to reduce redundancy is worth discussing. However, most of the existing band selection (BS) methods usually select bands according to the statistics or correlation, which neglect the spectral characteristics of the desired target and are not specially designed for target detection. Therefore, this article proposed a novel BS method, called constrained-target BS with subspace partition (CTSPBS), to select an optimal subset with low internal correlation and strong target representability for the target detection task. By using a specially designed subspace partition method based on correlation distance (CDSP), CTSPBS divides the hyperspectral bands into several unrelated subspaces. Then, according to certain constrained-target band prioritization (BP) criteria, the band with the highest priority in each subset is selected to form the optimal subset for a specific target. Correspondingly, two versions of the proposed method, minimum variance BS with CDSP (CDSP_MinV) and minimum variance BS with CDSP (CDSP_MaxV), are derived to implement CTSPBS. Extensive experiments on three public hyperspectral datasets demonstrate that the proposed method exhibit more robust and effective performance than several state-of-the-art methods. Finally, this article focuses on the difficulty of marine benthos detection in mariculture application and proves the feasibility of the proposed method.

Reversible data hiding (RDH) for JPEG images has attracted extensive attentions in recent years. However, since DCT is a de-correlation operation, it is difficult for JPEG images to employ the image redundancy for RDH design like uncompressed images. In this paper, considering better utilizing the special properties of quantized DCT coefficients, a new RDH scheme for JPEG images based on multiple two-dimensional (2D) histograms modification is proposed. Firstly, by combining every two nonzero alternating current (AC) coefficients of adjacent DCT blocks in each band as a pair, a new coefficients pairing strategy is proposed for a sharper 2D histogram. Then, with a classification process, multiple 1D and 2D histograms are generated, in which the 2D histograms are selected for data embedding and the bins in 1D histograms are shifted for reversibility. Finally, to further enhance the embedding performance, the 2D mappings are adaptively determined for different 2D histograms through a formulated rate-distortion model. Experimental results demonstrate the superiority of the proposed scheme both in visual quality and file size preservation.

Blood components such as hemoglobin, glucose, creatinine measuring are essential for monitoring one’s health condition. The current blood component measurement approaches still depend on invasive techniques that are painful, and uncomfortable for the patients. To facilitate measurement at home, we proposed a novel non-invasive technique to measure blood hemoglobin, glucose, and creatinine level based on PPG signal using Deep Neural Networks (DNN). Fingertip videos from 93 subjects have been collected using a smartphone. The PPG signal is generated from each video, and 46 characteristic features are then extracted from the PPG signal, its derivatives ( $1^{ \text {st}}$ and $2^{ \text {nd}}$ ) and from Fourier analysis. Additionally, age and gender are also included to feature because of the significant effects on hemoglobin, glucose, and creatinine. A correlation-based feature selection (CFS) using genetic algorithms (GA) has been used to select the optimal features to avoid redundancy and over-fitting. Finally, DNN based models have been developed to estimate the blood Hemoglobin (Hb), Glucose (Gl), and Creatinine (Cr) levels from the selected features. The approach provides the best-estimated accuracy of $R^{2} = 0.922$ for Hb, $R^{2} = 0.902$ for Gl, and $R^{2} = 0.969$ for Cr. Experimental aftermaths show that the proposed method is a suitable technique to be used clinically to measure human blood component levels without taking blood samples. This paper also reveals that smartphone-based PPG signal has a great potential to measure the different blood components.

Video images with measuring information in automated systems have a large volume. For example, the volume of video images of structures of natural origin is kilos and megabytes of information. Video images of natural structures have a high level of redundancy Reducing file volumes can be obtained through compression. Solving compression problems, or more generally, video image coding, uses advances and stimulates the development of many fields of engineering and mathematics. In the course of the research, a video compression using the structure of the zero-tree wavelet was selected. The results of the research have resulted in the development of existing wavelet compression techniques and the identification of optimal wavelet function for video images of this type. The best wavelet function for video images with measuring information in automated systems is Haar wavelet which characteristic indicators provide effective compression and fairly acceptable recovery. These indicators were the compression factor and the recovery error. The experimental results for five samples of video images of natural structures showed a pattern and a relationship between the appearance of the sample and its coefficients. You can see some equivalence of compression ratios, which indicates the similarity of video images, namely the identity of the outer surfaces of hard rock.

Granular structure is a mathematical expression of knowledge in granular computing and a direct determinant of the data processing efficiency. To improve the efficiency of data processing, many scholars have studied the reduction of granular structure. The attribute reduction and the granular reduction are two types of reduction on different layers of a granular structure, with the latter being both an essential step for granular structure reduction and the foundation of the attribute reduction. Yet compared with the attribute reduction, the granular reduction has received less attention from scholars. Therefore, a fuzzy granular reduction theory and a granular matrix based on the fuzzy $\beta$ -coverings is proposed in this article. The insufficiency of the existing granular reduction theory for fuzzy $\beta$ -coverings is pointed out, and proper sufficient and necessary conditions for two fuzzy $\beta$ -coverings generating the same upper and lower approximations are also given in this article. In addition, to reduce and evaluate a fuzzy $\beta$ -covering, a novel reduction algorithm based on a granular matrix is proposed for the first time. Also, since fuzzy covering reduction is NP-hard, a heuristic greedy algorithm is designed to obtain a reduct. Numerical experiments show that the redundancy rates of neighborhood granule sets induced by some big-scale data sets exceed 99 $\%$ , which indicates that the existing neighborhood granulation methods need to be urgently improved. Based on this, concise granular structures and much more efficient feature selection algorithms can be proposed in the future.

The performance of some three-dimensional mutual information-based algorithms can be affected, since only relevance and interaction are considered. Aiming at solving the problem, a feature selection algorithm based on redundancy analysis and interaction weight is proposed in this paper. The proposed algorithm adopts three-way interaction information to measure the interaction among the class label and features, and processes features for interaction weight analysis. Then, it employs symmetric uncertainty to measure the relevance between features and the class label as well as the redundancy between features, and selects the features with greater relevance and interaction as well as smaller redundancy. To validate the performance, the proposed algorithm is compared with several feature selection algorithms. Since relevance, redundancy, and interaction analysis are all presented, the proposed algorithm can obtain better feature selection performance.

Pole number recognition is highly important for the positioning tasks in high-speed rail catenary systems. The complicated working environment poses difficulties for number recognition algorithms, and this situation becomes even more challenging when illumination changes, image blurs and occlusions are considered. In this work, we present a high-accuracy pole number recognition framework including a high-performance cascaded CNN-based Detection and Recognition YOLO (DR-YOLO) and a temporal redundancy approach. First, the DR-YOLO utilizes global features for coarse plate detection and local features for accurate number recognition. Next, context-based combination of adjacent frames utilizes the complementarity and consistency of the same pole number recognition results and generates a unique result. The context-based amendment of adjacent poles utilizes the continuity of adjacent poles and corrects the fault or partial missing results caused by blur or occlusions. Extensive experimental testing is performed on 4 datasets representing high-speed train routine working environments. The reported experimental results validate the effectiveness and efficiency of the proposed approach.

Surveillance videos which record crowd behaviors have dramatically increased due to the wide applications. A quick view of such crowd surveillance video in a constrained time is an increasing demand because it always contain a huge number of redundancy frames. In this paper, we focus on summarization of crowd surveillance videos. But it is not easy due to two reasons. First, how to make the decision to keep or discard a subshot from the input surveillance video stream so that the summary can outline the main behaviors of the crowd over a limited frames sequence. Second, how to maintain performance of summarization model for long surveillance videos. To tackle these challenges, we formulate surveillance video summarization as a sequential decision-making process and train the summarization network with reinforcement learning-based framework. A novel crowd location-density reward is proposed to teach summarization network to produce high-quality summaries. In addition, a summarization network with three layers LSTM is designed to maintain performance across longer time spans. Extensive experiments on three public crowd surveillance videos datasets show that the proposed method achieves state-of-the-art performance.

The analysis of image processing technologies shows the main practice way to improve the quality of image processing. It is a preliminary analysis and subsequent image processing. It depends on the result of the preliminary analysis (filtration, sharpening, noise reduction, etc.). However, when selection of the method of preliminary analysis, an intermediate evaluation of results, selection of the subsequent processing method, etc. decision makers involved. This is not acceptable for practical implementation in automatic processing and transmission of video information systems. The main difficulties in working with video are large volumes of transmitted information and sensitivity to delays in the video information transmission. Therefore, in order to eliminate the maximum redundancy amount in the formation of the video sequence, 3 types of frames are used: I, P and B which form a frame group. Therefore, the possibility of upgrading coding methods for P-frames is considered on preliminary blocks' type identification with the subsequent formation of block code structures. As the correlation coefficient between adjacent frames increases, the compression ratio of the differential-represented frame's binary mask increases. A mechanism for processing predicted frames in the technology of compression of transformed images in computer systems and special purpose networks has been created. It based on the using of filter masks and the definition of complexity structural indicators of video fragments. It allows us to increase the efficiency of contours detection, namely, the accuracy of the allocation and localization of the semantic component up to 30% with an insignificant increase in the total processing time (no more than 5%).

This paper presents a new framework for wireless video streaming services using the resource-constrained embedded camera sensors based on block compressive sensing. We propose adaptive encoding scheme that exploits high redundancy between successive video frames to reduce transmission cost while effectively handling occlusion effects and, at the same time, maintains the simple and energy conserving encoder design. The proposed methodology adapts the compression ratio for different blocks of the non-key frame depending on temporal correlation. We also propose compressed domain motion and measurements estimation techniques to exploit the high correlation between successive frames at the decoder. The proposed motion estimation technique makes use of restricted isometry property of the sensing matrix to seek the best matching measurement vector for motion estimation as opposed to block matching in conventional video coding. In the proposed measurement estimation technique, efficient utilization of bandwidth is achieved by skipping some measurements at the transmitter side. The skipped measurements are estimated at the receiver by exploiting the correlation between CS measurements of the non-key frame and corresponding motion predicted frame using multiple regression model. Extensive simulation results on a set of diverse video sequences are presented to demonstrate the effectiveness of the proposed video coding technique.

Spectral analysis and Information-theory are two powerful and successful frameworks for feature selection in supervised classification problems. However, most of the methods developed under these frameworks have been introduced for handling exclusively numerical or non- numerical data. In this paper, we propose a supervised filter feature selection method that combines Spectral Feature Selection and Information-theory based redundancy analysis for selecting relevant and non-redundant features in supervised mixed datasets; i.e., datasets where the objects are described simultaneously by both, numerical and non-numerical features. To demonstrate the effectiveness of our proposed supervised filter feature selection method, we conducted several experiments on 40 public real-world datasets. Additionally, we compare our method against relevant state-of-the-art supervised filter methods for numerical, non-numerical, and mixed data. From this comparison, our method, in general, obtains better results than the results obtained by the other evaluated filter feature selection methods.

Contextual information can be used in recommender systems to make recommendation more efficient. Recent research has made progress in combining contextual information into representation models for recommendations. However, the existing approaches do not well address the problem of data sparsity, and they suffer from context redundancy. To deal with these problems, this paper proposes a context-aware recommendation approach based on embedded feature selection. It gets rid of context redundancy by generating a minimum subset of all contextual information and allocates the weight to each context appropriately. Experiments on the restaurant recommendation shows that the proposed approach has better performance.

Vision is the main way for people to perceive and recognize the world. In this paper, four categories of the redundant information of video encoding, spatial redundancy, time redundancy, visual redundancy and statistical redundancy, are first introduced. Spatial redundancy is redundant information existing in static frames, time redundancy is the luminance and chrominance correlation between adjacent frames in the video sequence, visual redundancy is information that the human visual system cannot perceive, statistical redundancy is the expression redundancy of the information source symbols. Then the performance assessment of video coding are introduced. And the future development of video coding and conclusion are finally discussed in this paper.

Unsupervised feature selection is an indispensable pre-processing step in many data mining and pattern recognition tasks where the unlabeled high dimensional data are ubiquitous. Most of existing methods fail to explore the local geometric structure consistency (preservation) of the input data and minimize redundancy of selected features simultaneously. In this paper we propose a novel unsupervised feature selection method which jointly integrates the local geometric structure consistency and redundancy minimization (JLSPRM) into an unified framework. JLSPRM utilizes nonnegative spectral analysis to learn the cluster labels of the input data, then the local geometric structure consistency is developed to make the learned cluster labels more accurate, during which the feature selection operation is performed. To minimize the redundancy rate among selected features, the maximal information coefficient (MIC) is utilized to evaluate the correlation of the pairwise features. Besides, the l2,1-norm is exerted on feature selection matrix which makes the framework decent for selecting features. An efficient iterative optimization algorithm is designed to obtain the solution of the unsupervised feature selection model. The superiority and effectiveness of our proposed approach over the state-of-the-art feature selection methods have also been validated through the extensive experiments on nine benchmark datasets.

Feature selection is an important data preprocessing method before classification. Multi-objective optimization algorithms have been proved an effective way to solve feature selection problems. However, there are few studies on multi-objective optimization feature selection methods for multi-label data. In this paper, a multi-objective multi-label filter feature selection algorithm based on two particle swarms (MOMFS) is proposed. We use mutual information to measure the relevance between features and label sets, and the redundancy between features, which are taken as two objectives. In order to avoid Particle Swarm Optimization (PSO) from falling into the local optimum and obtaining a false Pareto front, we employ two swarms to optimize the two objectives separately and propose an improved hybrid topology based on particle’s fitness value. Furthermore, an archive maintenance strategy is introduced to maintain the distribution of archive. In order to study the effectiveness of the proposed algorithm, we select five multi-label evaluation criteria and perform experiments on seven multi-label data sets. MOMFS is compared with classic single-objective multi-label feature selection algorithms, multi-objective filter and wrapper feature selection algorithms. The experimental results show that MOMFS can effectively reduce the multi-label data dimension and perform better than other approaches on five evaluation criteria.

Feature selection plays a fundamental role in many data mining and machine learning tasks. In this paper, we proposed a novel feature selection method, namely, Dynamic Feature Selection Method with Minimum Redundancy Information (MRIDFS). In MRIDFS, the conditional mutual information is used to calculate the relevance and the redundancy among multiple features, and a new concept, the feature-dependent redundancy ratio, was introduced. Such ratio can represent redundancy more accurately. To evaluate our method, MRIDFS is tested and compared with seven popular methods on 16 benchmark data sets. Experimental results show that MRIDFS outperforms in terms of average classification accuracy.

In this paper, we focus on various compression and classification algorithms for three different paralinguistic signal classification tasks. These tasks are quite difficult for humans because the sound information from such signals is difficult to distinguish. Therefore, when machine learning techniques are applied to analyze paralinguistic signals, several different aspects of speech-related information, such as prosody, energy, and cepstral information, are usually considered for feature extraction. However, when the size of the training corpus is not sufficiently large, it is extremely difficult to directly apply machine learning to classify such signals due to their high feature dimensions; this problem is also known as the curse of dimensionality. This paper proposes to address this limitation by means of feature compression. First, we present experimental results obtained by using various compression algorithms to compress signals to eliminate redundancy of the signal features. We observe that compared with the original features, the compressed signal features still provide a comparable ability to distinguish the signals, especially when using a fully connected neural network classifier. Second, we calculate the output distribution of the F1-score for each emotion in the speech emotion recognition problem and show that the fully connected neural network classifier performs more stably than other classical methods.

In this paper, a computer based system has been proposed as a support to gastrointestinal polyp detection. It can detect and classify gastrointestinal polyps from endoscopic video. Color wavelet (CW) features and convolutional neural network (CNN) features of endoscopic video frames are extracted. Mutual information based feature selection technique-Minimum redundancy maximum relevance (mRMR) is used to scale down feature vector. Instead of using a single classifier, Bootstrap Aggregrating (Bagging)- an ensemble classifier is used. Proposed system has been assessed against different public databases and our own datasets. Evaluation shows that, the system outperforms the existing methods.

Intra prediction is an effective method for video coding to remove the spatial redundancy of content. Classical intra prediction method usually creates a prediction block by extrapolating the encoded pixels surrounding the target block. However, existing methods cannot guarantee the prediction efficiency for rich textural structure, especially when weak spatial correlation exists between the target block and reference pixels. To remedy this issue, this paper proposes a novel intra prediction method via convolutional encoder-decoder network, which we term IPCED. IPCED can learn and extract the internal representation of reference blocks, and progressively generate a prediction block from this representation. IPCED is a data-driven method, which represents an improvement over hand-crafted methods, and is capable of improving the accuracy of intra prediction. Extensive experimental results demonstrate that IPCED can generate higher-quality intra prediction results, achieves 3.41%, 3.07% and 3.44% bitrate saving for the Y/Cb/Cr channel compared with HEVC baseline, which is significantly beyond existing methods.

The wireless visual sensor network (WVSN) as a new emerged intelligent visual system, has been applied in many video monitoring sites. However, there is still great challenge because of the limited wireless network bandwidth. To resolve the problem, we propose a real-time dynamic texture approach which can detect and reduce the temporal redundancy during many successive image frames. Firstly, an adaptively learning background model is improved to discover successive similar image frames from the inputting video sequence. Then, the dynamic texture model based on the singular value decomposition is adopted to distinguish foreground and background element dynamics. Furthermore, a background discarding strategy based on visual motion coherence is proposed to determine whether each image frame is streamed or not. To evaluate the trade-off performance of the proposed method, it is tested on the CDW-2014 dataset, which can accurately detect the first foreground frame when the moving objects of interest appear in the field of view in the most tested dynamic scenes, and the misdetection rate of the undetected foreground frames is near to zero. Compared to the original stream, it can reduce the occupied bandwidth a lot and its computational cost is relatively lower than the state-of-the-art methods.

The forensics of the median filtering is a challenging task due to its content preserving nature. Several methods have been proposed for median filtering forensics in digital images. However the performance of these methods deteriorates for compressed images, small resolutions of images and for anti-forensic operations. Moreover large feature set dimensions of these methods also pose a computational challenge. This paper proposes, a 8-dimensional feature set, derived from two state-of-the-art techniques by employing maximum relevancy and minimum redundancy (mRMR) feature selection approach. Features are selected by mRMR on the basis of distance correlation as an association measure. Extensive experiments are performed to evaluate the efficacy of proposed method through six different databases. The proposed method outperforms state-of-the-art techniques for uncompressed images, compressed images at low quality factors, low resolutions images and for an anti-forensic operation. The performance of the proposed method is also compared with convolutional neural network (CNN) based features for the detection of median filtering at low resolutions and for compressed images. Also, experimental results support the performance of proposed method over other manipulations (average and Gaussian filtering).

High dimensionality and sample imbalance of gene expression data promote the development of effective algorithms for classifying gene expression data. To improve the ability to distinguish different subtypes of gene expression data, we devise a hypervolume-based discrete evolutionary optimization algorithm (HYBDEOA) in this paper. Four objectives, namely the number of genes, the accuracy, the relevance, and the redundancy, are optimized simultaneously to guide the evolution. Firstly, binary encoding is used to choose some features, projecting data onto different subspaces. After that, a discrete neighborhood operation is conducted to generate a new binary-mapped population. Combining the new population with the current population, we employ the hypervolume-based mechanism to select the Pareto solutions. Finally, a discrete mutation method is proposed to find promising solutions in the binary search space. To demonstrate the performance of HYBDEOA, we apply HYBDEOA to 55 synthetic datasets and 35 cancer gene expression datasets. Extensive experiments are also conducted to reveal the effectiveness and efficiency of HYBDEOA. The experimental results demonstrate that our proposed method is a parameter-less and robust algorithm, which can group gene expression data with a finer and more informative classification.

Feature selection is an effective way to reduce the data dimensionality of hyperspectral imagery and obtain a better performance in the subsequent applications, such as classification. The ideal approach is to obtain the optimal tradeoff between two criteria for hyperspectral image feature selection: 1) information preservation and 2) redundancy reduction. However, constructing a hyperspectral feature selection model for the above two criteria is difficult due to the complexity of hyperspectral imagery. Although evolutionary multiobjective optimization methods have been recently presented to simultaneously optimize the above criteria, they cannot control the global exploration versus local exploitation capabilities in the search space for the hyperspectral feature selection problem. Thus, in this article, a novel discrete sine cosine algorithm (SCA)-based multiobjective feature selection (MOSCA_FS) approach is proposed for hyperspectral imagery. In the proposed method, a novel and effective framework of multiobjective hyperspectral feature selection is designed. In the framework, the ratio between the Jeffries–Matusita (JM) distance and mutual information (MI) is modeled to minimize the redundancy and maximize the relevance of the selected feature subset. In addition, another measurement—the variance (Var)—is applied for maximizing the information amount. Furthermore, to resolve the discrete hyperspectral feature selection problem, a novel discrete SCA is first proposed, which enhances the selection of the ideal feature subset. The effectiveness and universality of the proposed method was verified by experiments with ten University of California at Irvine (UCI) data sets, five hyperspectral image data sets, and one spectral data set of typical surface features.

Approaches to image compression with machine learning now achieve superior performance on the compression rate compared to existing hybrid codecs. The conventional learning-based methods for image compression exploits hyper-prior and spatial context model to facilitate probability estimations. Such models have limitations in modeling long-term dependency and do not fully squeeze out the spatial redundancy in images. In this paper, we propose a coarse-to-fine framework with hierarchical layers of hyper-priors to conduct comprehensive analysis of the image and more effectively reduce spatial redundancy, which improves the rate-distortion performance of image compression significantly. Signal Preserving Hyper Transforms are designed to achieve an in-depth analysis of the latent representation and the Information Aggregation Reconstruction sub-network is proposed to maximally utilize side-information for reconstruction. Experimental results show the effectiveness of the proposed network to efficiently reduce the redundancies in images and improve the rate-distortion performance, especially for high-resolution images. Our project is publicly available at https://huzi96.github.io/coarse-to-fine-compression.html.

Motivation The Protein Data Bank (PDB), the ultimate source for data in structural biology, is inherently imbalanced. To alleviate biases, virtually all structural biology studies use nonredundant (NR) subsets of the PDB, which include only a fraction of the available data. An alternative approach, dubbed redundancy-weighting (RW), down-weights redundant entries rather than discarding them. This approach may be particularly helpful for machine-learning (ML) methods that use the PDB as their source for data. Methods for secondary structure prediction (SSP) have greatly improved over the years with recent studies achieving above 70% accuracy for eight-class (DSSP) prediction. As these methods typically incorporate ML techniques, training on RW datasets might improve accuracy, as well as pave the way toward larger and more informative secondary structure classes. Results This study compares the SSP performances of deep-learning models trained on either RW or NR datasets. We show that training on RW sets consistently results in better prediction of 3- (HCE), 8- (DSSP) and 13-class (STR2) secondary structures. Availability and implementation The ML models, the datasets used for their derivation and testing, and a stand-alone SSP program for DSSP and STR2 predictions, are freely available under LGPL license in http://meshi1.cs.bgu.ac.il/rw. Supplementary information Supplementary data are available at Bioinformatics online.

In this letter, we propose a concise feature representation framework for acoustic scene classification by pruning embeddings obtained from SoundNet, a deep convolutional neural network. We demonstrate that the feature maps generated at various layers of SoundNet have redundancy. The proposed singular value decomposition based method reduces the redundancy while relying on the assumption that useful feature maps produced by different classes lie along different directions. This leads to ignoring the feature maps that produce similar embeddings for different classes. In the context of using an ensemble of classifiers on the various layers of SoundNet, pruning the redundant feature maps leads to reduction in dimensionality and computational complexity. Our experiments on acoustic scene classification demonstrate that ignoring 73% of feature maps reduces the performance by less than 1% with 12.67% reduction in computational complexity. In addition to this, we also show that the proposed pruning framework can be utilized to remove filters in the SoundNet network architecture, with 13x lesser model storage requirement. Also, the number of parameters reduce from 28 million to 2 million with marginal degradation in performance. This small model obtained after applying the proposed pruning procedure is evaluated on different acoustic scene classification datasets, and shows excellent generalization ability.