Robust Features Research Articles

The voice of depression: speech features as biomarkers for major depressive disorder

BackgroundPsychiatry faces a challenge due to the lack of objective biomarkers, as current assessments are based on subjective evaluations. Automated speech analysis shows promise in detecting symptom severity in depressed patients. This project aimed to identify discriminating speech features between patients with major depressive disorder (MDD) and healthy controls (HCs) by examining associations with symptom severity measures.MethodsForty-four MDD patients from the Psychiatry Department, University Hospital Aachen, Germany and fifty-two HCs were recruited. Participants described positive and negative life events, which were recorded for analysis. The Beck Depression Inventory (BDI-II) and the Hamilton Rating Scale for Depression gauged depression severity. Transcribed audio recordings underwent feature extraction, including acoustics, speech rate, and content. Machine learning models including speech features and neuropsychological assessments, were used to differentiate between the MDD patients and HCs.ResultsAcoustic variables such as pitch and loudness differed significantly between the MDD patients and HCs (effect sizes \U0001d73c2 between 0.183 and 0.3, p < 0.001). Furthermore, variables pertaining to temporality, lexical richness, and speech sentiment displayed moderate to high effect sizes (\U0001d73c2 between 0.062 and 0.143, p < 0.02). A support vector machine (SVM) model based on 10 acoustic features showed a high performance (AUC = 0.93) in differentiating between HCs and patients with MDD, comparable to an SVM based on the BDI-II (AUC = 0.99, p = 0.01).ConclusionsThis study identified robust speech features associated with MDD. A machine learning model based on speech features yielded similar results to an established pen-and-paper depression assessment. In the future, these findings may shape voice-based biomarkers, enhancing clinical diagnosis and MDD monitoring.

Complete suspension culture of human induced pluripotent stem cells supplemented with suppressors of spontaneous differentiation

Human induced pluripotent stem cells (hiPSCs) are promising resources for producing various types of tissues in regenerative medicine; however, the improvement in a scalable culture system that can precisely control the cellular status of hiPSCs is needed. Utilizing suspension culture without microcarriers or special materials allows for massive production, automation, cost-effectiveness, and safety assurance in industrialized regenerative medicine. Here, we found that hiPSCs cultured in suspension conditions with continuous agitation without microcarriers or extracellular matrix components were more prone to spontaneous differentiation than those cultured in conventional adherent conditions. Adding PKCβ and Wnt signaling pathway inhibitors in the suspension conditions suppressed the spontaneous differentiation of hiPSCs into ectoderm and mesendoderm, respectively. In these conditions, we successfully completed the culture processes of hiPSCs, including the generation of hiPSCs from peripheral blood mononuclear cells with the expansion of bulk population and single-cell sorted clones, long-term culture with robust self-renewal characteristics, single-cell cloning, direct cryopreservation from suspension culture and their successful recovery, and efficient mass production of a clinical-grade hiPSC line. Our results demonstrate that precise control of the cellular status in suspension culture conditions paves the way for their stable and automated clinical application.

Complete suspension culture of human induced pluripotent stem cells supplemented with suppressors of spontaneous differentiation.

Human induced pluripotent stem cells (hiPSCs) are promising resources for producing various types of tissues in regenerative medicine; however, the improvement in a scalable culture system that can precisely control the cellular status of hiPSCs is needed. Utilizing suspension culture without microcarriers or special materials allows for massive production, automation, cost-effectiveness, and safety assurance in industrialized regenerative medicine. Here, we found that hiPSCs cultured in suspension conditions with continuous agitation without microcarriers or extracellular matrix components were more prone to spontaneous differentiation than those cultured in conventional adherent conditions. Adding PKCβ and Wnt signaling pathway inhibitors in the suspension conditions suppressed the spontaneous differentiation of hiPSCs into ectoderm and mesendoderm, respectively. In these conditions, we successfully completed the culture processes of hiPSCs, including the generation of hiPSCs from peripheral blood mononuclear cells with the expansion of bulk population and single-cell sorted clones, long-term culture with robust self-renewal characteristics, single-cell cloning, direct cryopreservation from suspension culture and their successful recovery, and efficient mass production of a clinical-grade hiPSC line. Our results demonstrate that precise control of the cellular status in suspension culture conditions paves the way for their stable and automated clinical application.

Trans-retinal predictive signals of visual features are precise, saccade-specific and operate over a wide range of spatial frequencies.

Saccadic eye movements successively project the saccade target on two retinal locations: a peripheral one before the saccade, and the fovea after the saccade. Typically, performance in discriminating stimulus features changes between these two projections is very poor. However, a short (~ 200 ms) blanking of the target upon saccade onset drastically improves performance, demonstrating that a precise signal of the peripheral projection is retained during the saccade. While little is known about the nature of that transsaccadic signal, previous reports conjectured that it relies on information processed by the magnocellular system. Across two experiments, we investigated the feature blanking effect for a wide range of spatial frequencies (0.5 to 8 cycles per degree of visual angle, dva), stimulus sizes (1 to 4 dva) and eccentricities (6 to 10 dva). In each trial, participants executed a saccade to a high-contrast grating presented either left or right of fixation. During the saccade, the grating changed orientation (clockwise or counter-clockwise) either instantaneously or after a 200 ms blank, and participants reported the change's direction. We contrasted this saccade condition with a trans-retinal fixation condition mimicking the peripheral-then-foveal sequence of the target stimulus occurring across a saccade. Remarkably, blanking improved performance reliably for each spatial frequency, stimulus size, and eccentricity, but only in the saccade condition. Performance with blanking in saccade trials systematically exceeded performance in the fixation condition. Our results demonstrate a robust feature blanking effect across saccades, suggesting that transsaccadic processes involve low-level visual features beyond those processed in the magnocellular system.

CNSC-65. ELECTROPHYSIOLOGICAL AND SPATIAL TRANSCRIPTOMIC PROFILING OF GLIOMA-INFILTRATED CORTEX REVEAL LAYER SPECIFIC ALTERATIONS

Abstract Specialized neocortical circuits selectively organize neuronal signals and encode features of cognitive processing via local and long-range connections. Since glioma-infiltrated cortex is excitable and can participate in cognitive processing, the underlying laminar structure and functionality may still be preserved despite infiltration. Moreover, since glioma cells remodel existing neural circuits and microenvironmental factors drive invasion and proliferation, neuron-glioma interactions may demonstrate layer specificity. As glioma-infiltrated cortex remains poorly understood, we sought to characterize these regions using electrophysiological, structural, and genomic approaches. We assessed the power spectra of normal-appearing and glioma-infiltrated cortex using magnetoencephalography and subdural high-density electrode arrays. Immunohistochemistry of formalin-fixed paraffin-embedded (FFPE) samples of infiltrated cortex enabled protein-level neuronal and glioma identification to determine laminar preservation and spatial invasion patterns. Spatial transcriptomics of FFPE tissues and single-nucleus RNA-sequencing were used to identify cell populations, genomic alterations, and cell-cell communication within and across samples at testing and validation sites across the US and Europe. Increased delta range (1-4 Hz) power and decreased power in the beta range (12-20 Hz), activity thought to originate from deep layers, were identified as robust features of infiltrated cortex which were maintained across glioma subtypes (n = 139 patients) and preserved in a validation cohort (n = 8 patients). Immunohistochemistry revealed greater tumor burden within deep laminae regardless of glioma subtype (n = 20 samples). Tissue proteomics and spatial genomics analyses performed using glioma-infiltrated cortical samples (n = 10 samples; 29,011 spots) confirmed routine preservation of laminar organization, greater tumor burden in deep layers, as well as layer specific differences in glioma-related expression programs. Cell-cell communication analyses demonstrated increased interactions in glioma-infiltrated cortex across layers. These findings suggest that laminar structure may be preserved in glioma-infiltrated cortex while remodeling of neural circuits alters spatiotemporal activity in a predictable manner and support a deep to superficial invasion pattern. Thus, investigating glioma infiltration using layer-centric approaches may facilitate novel insights into neuron-glioma interactions.

NIMG-42. ANALYSIS OF TUMOR RELAPSE PROBABILITY AND OVERALL SURVIVAL PREDICTION FOLLOWING TEMOZOLOMIDE CHEMORADIATION USING FET PET INCLUDING RADIOMICS IN PATIENTS WITH GLIOBLASTOMA

Abstract BACKGROUND Early after surgery and completion of first-line concomitant chemoradiation with temozolomide, the evaluation of tumor relapse probability and overall survival prediction is of considerable interest for the management of patients with glioblastoma METHODS Sixty-three newly diagnosed patients with glioblastoma (age range, 19-82) who received static and dynamic PET imaging using the radiolabeled amino acid O-(2-[18F]fluoroethyl)-L-tyrosine (FET) after surgery or biopsy and completion of concomitant chemoradiation with temozolomide were evaluated. Static FET PET parameters such as maximum and mean tumor-to-brain ratios (TBRmax, TBRmean) and metabolic tumor volumes (MTV), and the dynamic FET PET parameter time-to-peak (TTP) were obtained. Additionally, FET PET radiomics features (n=1,303) were extracted for each patient and a test-retest analysis was performed to identify robust features prior to feature selection. The prognostic value of FET PET parameters and radiomics features was evaluated using receiver-operating-characteristic (ROC) analyses with regard to a significantly prolonged progression-free and overall survival (PFS, OS). Subsequently, univariate and multivariate survival estimates were performed to assess the prognostic value of these parameters to predict a significantly longer PFS and OS RESULTS ROC analyses revealed that the parameter TBRmax was a powerful parameter to predict both a significantly longer PFS (20.3 vs. 8.9 months; P=0.002; threshold, 2.75) and OS (34.9 vs. 18.7 months; P=0.005; threshold, 2.75). MTV had a similar prognostic value for both PFS (16.7 vs. 7.2 months; P=0.002; threshold, 34.0 mL) and OS (30.6 vs. 14.8 months; P=0.002; threshold, 32.6 mL). After feature selection, 3 of 15 selected radiomics features predicted a significantly longer PFS and OS in univariate analyses. TBRmax, MTV, and one of 15 radiomics features remained significant in multivariate survival analysis (all P≤0.03) CONCLUSION Our results suggest that FET PET and radiomics parameters were highly prognostic in this group of patients at an early stage of first-line therapy.

A Quad‐Port MIMO Antenna With Improved Isolation Developed by Colored Resin Fiber Material for Wideband Applications

ABSTRACTThe article explores a quad‐port compact (56 × 56 × 1.5 mm3) multiple‐input multiple‐output (MIMO) antenna developed on a colored resin fiber material encompassing enhanced isolation for wideband applications. The design antenna configuration incorporates four ground stubs positioned among antenna pairs on the upper section of the substrate and eight rectangular‐shaped ground slots to enhance antenna isolation. The antenna functions over a broad spectral range from 4.21 to 10.95 GHz, providing coverage for WLAN (5.15–5.35 GHz/5.725–5.825 GHz), C band uplink (5.925–6.425 GHz), downlink defense (7.2–7.7 GHz), and ITU band (8–8.5 GHz), with 88.91% fractional bandwidth. The antenna design ensures exceptional isolation, exceeding 21 dB among orthogonal antennas and 25 dB between oppositely placed antennas across the entire functioning bandwidth. Across the working band, the antenna showcases favorable diversity characteristics, including a lower envelope correlation coefficient (ECC &lt; 0.085), higher diversity gain (DG &gt; 9.95 dB), minimal channel capacity loss (&lt; 0.39 bits/s/Hz), and negligible magnitude variation in mean effective gain among the antenna ports (&lt; 0.1 dB). The acquired total active reflection (TARC) curve exhibits the antenna's robust resonance features over the application band and commendable time domain properties in the active spectral region. The findings of the simulation and measurements revealed a close alignment.

Mf-net: multi-feature fusion network based on two-stream extraction and multi-scale enhancement for face forgery detection

Due to the increasing sophistication of face forgery techniques, the images generated are becoming more and more realistic and difficult for human eyes to distinguish. These face forgery techniques can cause problems such as fraud and social engineering attacks in facial recognition and identity verification areas. Therefore, researchers have worked on face forgery detection studies and have made significant progress. Current face forgery detection algorithms achieve high detection accuracy within-dataset. However, it is difficult to achieve satisfactory generalization performance in cross-dataset scenarios. In order to improve the cross-dataset detection performance of the model, this paper proposes a multi-feature fusion network based on two-stream extraction and multi-scale enhancement. First, we design a two-stream feature extraction module to obtain richer feature information. Secondly, the multi-scale feature enhancement module is proposed to focus the model more on information related to the current sub-region from different scales. Finally, the forgery detection module calculates the overlap between the features of the input image and real images during the training phase to determine the forgery regions. The method encourages the model to mine forgery features and learns generic and robust features not limited to a particular feature. Thus, the model achieves high detection accuracy and performance. We achieve the AUC of 99.70% and 90.71% on FaceForensics++ and WildDeepfake datasets. The generalization experiments on Celeb-DF-v2 and WildDeepfake datasets achieve the AUC of 80.16% and 65.15%. Comparison experiments with multiple methods on other benchmark datasets confirm the superior generalization performance of our proposed method while ensuring model detection accuracy. Our code can be found at: https://github.com/1241128239/MFNet.

Adaptive Depth Estimation via SoftHebbLayer in a Hybrid ResNet-UNet Architecture

Abstract. Depth estimation is a key task in computer vision, critical for applications such as autonomous navigation and augmented reality. This paper introduces a novel hybrid neural network that combines ResNet and UNet architectures with a SoftHebbLayer, inspired by Hebbian learning principles, to improve depth estimation from RGB images. The ResNet backbone extracts robust hierarchical features, while the UNet decoder reconstructs fine-grained depth maps. The SoftHebbLayer dynamically adjusts feature connections based on co-activation, enhancing the models adaptability to diverse environments. This approach addresses common challenges in depth estimation, including poor generalization and computational inefficiency. We evaluated the model on the DIODE dataset, achieving strong results in Mean Squared Error (MSE), which is 0.0800 and Root Mean Squared Error (RMSE), which is 0.2805, demonstrating improved accuracy in both indoor and outdoor scenes. While the model excels in precision, further refinement is needed to reduce computational overhead and improve performance in challenging environments. This research paves the way for more efficient, adaptable depth estimation models, with potential applications in mobile robotics and real-time edge computing systems.

IShapEditing: Intelligent Shape Editing with Diffusion Models

AbstractRecent advancements in generative models have enabled image editing very effective with impressive results. By extending this progress to 3D geometry models, we introduce iShapEditing, a novel framework for 3D shape editing which is applicable to both generated and real shapes. Users manipulate shapes by dragging handle points to corresponding targets, offering an intuitive and intelligent editing interface. Leveraging the Triplane Diffusion model and robust intermediate feature correspondence, our framework utilizes classifier guidance to adjust noise representations during sampling process, ensuring alignment with user expectations while preserving plausibility. For real shapes, we employ shape predictions at each time step alongside a DDPM‐based inversion algorithm to derive their latent codes, facilitating seamless editing. iShapEditing provides effective and intelligent control over shapes without the need for additional model training or fine‐tuning. Experimental examples demonstrate the effectiveness and superiority of our method in terms of editing accuracy and plausibility.

Multi-task deep learning for quantifying methane emissions from 2-D plume imagery with Low Signal-to-Noise Ratio

ABSTRACT Methane is the second most significant greenhouse gas after carbon dioxide. As global warming intensifies, the quantification of methane point sources is becoming increasingly crucial. However, retrieving high-quality methane signals from remote sensing data remains challenging due to various factors, including surface reflectance, atmospheric interference, sensor noise, wind speed, and sensor sensitivity. In real-world scenarios, methane remote sensing quantification often encounters unfavourable conditions that lead to low signal-to-noise ratio (SNR) signals, resulting in reduced quantification accuracy. To address these challenges, we introduce a novel multi-task learning-based approach. Specifically, we incorporate a denoising auxiliary task into the quantification network by introducing an additional denoising branch that recovers clean plume column concentration maps. The network is trained with supervision using both denoising and quantification losses, which enables it to acquire robust feature representations to noise and benefits the quantification of low SNR images. During the inference phase, the denoising branch is removed, resulting in an efficient and robust single quantification network with reduced inferring time, supporting onboard computation. We construct a low SNR database based on the AVIRIS-NG sensor and evaluate the generalization ability of our method. DQNet achieves the highest RMSE, MAPE, and R of 16.478 kg/h, 15.296%, and 95.167% respectively on the synthetic dataset. Across the entire range of SNR, DQNet exhibits a greater relative advantage as SNR decreases. However, our approach is not limited to AVIRIS-NG and can be easily extended to various multispectral and hyperspectral satellites.

Enhanced Tuberculosis Detection in Chest X-Rays Using Optimal Feature Selection with Hybrid Binary Particle Swarm Optimization (HBPSO) and a Dual Classifier Framework Combining LSTM and Spiking Neural Networks

Tuberculosis (TB) remains a significant global health challenge, particularly in developing countries, where rapid and accurate diagnosis is crucial for effective treatment. Recent advancements in machine learning and image processing have shown promise in enhancing the detection of TB from chest X-ray images. This research introduces a novel approach that integrates a hybrid feature selection method with a dual classifier framework to improve the accuracy of tuberculosis detection. Feature extraction is conducted using Multi-Scale Local Binary Patterns (MSLBP), Speeded-Up Robust Features (SURF) with Bag of Words (BoW), and Cartoon Texture Features, combining them into a comprehensive feature vector. The Hybrid Binary Particle Swarm Optimization (HBPSO) is employed for optimal feature selection, reducing redundancy and improving classification efficiency. Classification is performed using a dual framework involving a Long Short-Term Memory (LSTM) network and a Spiking Neural Network (SNN), which work in tandem to provide robust predictions. Majority voting is utilized to finalize predictions, ensuring high accuracy and adaptability across varying TB presentations. Extensive evaluation on publicly available datasets shows that this hybrid approach significantly outperforms existing methods, achieving a highest accuracy of 99.73%, along with superior precision, recall, and F1 scores.

Clean-Label Backdoor Watermarking for Dataset Copyright Protection via Trigger Optimization

High-quality datasets are essential for training high-performance models, while the process of collection, cleaning, and labeling is costly. As a result, datasets are considered valuable intellectual property. However, when security mechanisms are symmetry-breaking, creating exploitable vulnerabilities, unauthorized use or data leakage can infringe on the copyright of dataset owners. In this study, we design a method to mount clean-label dataset watermarking based on trigger optimization, aiming to protect the copyright of the dataset from infringement. We first perform iterative optimization of the trigger based on a surrogate model, with targets class samples guiding the updates. The process ensures that the optimized triggers contain robust feature representations of the watermark target class. A watermarked dataset is obtained by embedding optimized triggers into randomly selected samples from the watermark target class. If an adversary trains a model with the watermarked dataset, our watermark will manipulate the model’s output. By observing the output of the suspect model on samples with triggers, it can be determined whether the model was trained on the watermarked dataset. The experimental results demonstrate that the proposed method exhibits high imperceptibility and strong robustness against pruning and fine-tuning attacks. Compared to existing methods, the proposed method significantly improves effectiveness at very low watermarking rates.

A robust multi-label feature selection based on label significance and fuzzy entropy

Multi-label feature selection is one of the key steps in dealing with multi-label classification problems in high-dimensional data. In this step, label enhancement techniques play an important role. However, it is worth noting that many current methods tend to ignore the intrinsic connection between inter-sample similarity and inter-label correlation when implementing label enhancement learning. The neglect may prevent the process of label enhancement from accurately revealing the complex structure and underlying patterns within data. For this reason, a fuzzy multi-label feature selection method based on label significance and fuzzy entropy is proposed. An innovative label enhancement technique that considers not only the intrinsic connection between features and labels, but also the correlation between labels was first devised. Based on this enhanced label representation, the concept of fuzzy entropy is further defined to quantify the uncertainty of features for multi-label classification tasks. Subsequently, a feature selection algorithm based on feature importance and label importance was developed. The algorithm guides the feature selection process by evaluating how much each feature contributes to multi-label classification and how important each label is to the overall classification task. Finally, through a series of experimental validation, the proposed algorithm is proved to have better performance for multi-label classification tasks.

CFRNet: Cross-Attention-Based Fusion and Refinement Network for Enhanced RGB-T Salient Object Detection.

Existing deep learning-based RGB-T salient object detection methods often struggle with effectively fusing RGB and thermal features. Therefore, obtaining high-quality features and fully integrating these two modalities are central research focuses. We developed an illumination prior-based coefficient predictor (MICP) to determine optimal interaction weights. We then designed a saliency-guided encoder (SG Encoder) to extract multi-scale thermal features incorporating saliency information. The SG Encoder guides the extraction of thermal features by leveraging their correlation with RGB features, particularly those with strong semantic relationships to salient object detection tasks. Finally, we employed a Cross-attention-based Fusion and Refinement Module (CrossFRM) to refine the fused features. The robust thermal features help refine the spatial focus of the fused features, aligning them more closely with salient objects. Experimental results demonstrate that our proposed approach can more accurately locate salient objects, significantly improving performance compared to 11 state-of-the-art methods.

Robust load feature extraction based secondary VMD novel short-term load demand forecasting framework

Load forecasting, as a crucial component of the electricity market, plays a significant role in ensuring the secure operation and rational planning of the power grid. However, as the power system becomes increasingly intricate, the demands on load forecasting techniques have escalated. Consequently, to mitigate the errors in short-term load forecasting (STLF) caused by uncertainty factors and to accommodate daily forecasting under abnormal electricity load conditions, this paper proposes a hybrid load forecasting model that combines an improved Secondary Variational Mode Decomposition (SVMD) algorithm with the Informer model. Employing electricity load data from the Panama context, the data is divided into four distinct experimental cases. The outcomes manifest that in contrast to the baseline model, the proposed approach engenders a minimal reduction of 15.08%, 12.95%, and 13.21% in Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE), and Root Mean Square Error (RMSE), respectively. Furthermore, supplementary experimental results demonstrate that the model exhibits strong robustness.

Optimized 3D-2D CNN for automatic mineral classification in hyperspectral images

Abstract Mineral classification using hyperspectral imaging represents an essential field of research improving the understanding of geological compositions. This study presents an advancedmethodology that uses an optimized 3D-2D CNNmodel for automatic mineral identification and classification. Our approach includes such crucial steps as using the Diagnostic Absorption Band (DAB) selection technique to selectively extract bands that contain the absorption features of minerals for classification in the Cuprite zone. Focusing on the Cuprite dataset, our study successfully identified the following minerals: alunite, calcite, chalcedony, halloysite, kaolinite,montmorillonite,muscovite, and nontronite. The Cuprite dataset results with an overall accuracy rate of 95.73%underscore the effectiveness of our approach and a significant improvement over the benchmarks established by related studies. Specifically, ASMLP achieved a 94.67%accuracy rate, followed by 3D CNN at 93.86%, SAI-MLP at 91.03%, RNN at 89.09%, SPE-MLP at 85.53%, and SAMat 83.31 %. Beyond the precise identification of specific minerals, ourmethodology proves its versatility for broader applications in hyperspectral image analysis. The optimized 3D-2D CNNmodel excels in terms of mineral identification and sets a new standard for robust feature extraction and classification.

Comprehensive empirical evaluation of feature extractors in computer vision

Feature detection and matching are fundamental components in computer vision, underpinning a broad spectrum of applications. This study offers a comprehensive evaluation of traditional feature detections and descriptors, analyzing methods such as Scale Invariant Feature Transform (SIFT), Speeded-Up Robust Features (SURF), Binary Robust Independent Elementary Features (BRIEF), Oriented FAST and Rotated BRIEF (ORB), Binary Robust Invariant Scalable Keypoints (BRISK), KAZE, Accelerated KAZE (AKAZE), Fast Retina Keypoint (FREAK), Dense and Accurate Invariant Scalable descriptor for Yale (DAISY), Features from Accelerated Segment Test (FAST), and STAR. Each feature extractor was assessed based on its architectural design and complexity, focusing on how these factors influence computational efficiency and robustness under various transformations. Utilizing the Image Matching Challenge Photo Tourism 2020 dataset, which includes over 1.5 million images, the study identifies the FAST algorithm as the most efficient detector when paired with the ORB descriptor and Brute-Force (BF) matcher, offering the fastest feature extraction and matching process. ORB is notably effective on affine-transformed and brightened images, while AKAZE excels in conditions involving blurring, fisheye distortion, image rotation, and perspective distortions. Through more than 2 million comparisons, the study highlights the feature extractors that demonstrate superior resilience across various conditions, including rotation, scaling, blurring, brightening, affine transformations, perspective distortions, fisheye distortion, and salt-and-pepper noise.

Color image watermarking using vector SNCM-HMT

An image watermarking scheme is typically evaluated using three main conflicting characteristics: imperceptibility, robustness, and capacity. Developing a good image watermarking method is challenging because it requires a trade-off between these three basic characteristics. In this paper, we proposed a statistical color image watermarking based on robust discrete nonseparable Shearlet transform (DNST)-fast quaternion generic polar complex exponential transform (FQGPCET) magnitude and vector skew-normal-Cauchy mixtures (SNCM)-hidden Markov tree (HMT). The proposed watermarking system consists of two main parts: watermark inserting and watermark extraction. In watermark inserting, we first perform DNST on R, G, and B components of color host image, respectively. We then compute block FQGPCET of DNST domain color components, and embed watermark signal in DNST-FQGPCET magnitudes using multiplicative approach. In watermark extraction, we first analyze the robustness and statistical characteristics of local DNST-FQGPCET magnitudes of color image. We then observe that, vector SNCM-HMT model can capture accurately the marginal distribution and multiple strong dependencies of local DNST-FQGPCET magnitudes. Meanwhile, vector SNCM-HMT parameters can be computed effectively using variational expectation–maximization (VEM) parameter estimation. Motivated by our modeling results, we finally develop a new statistical color image watermark decoder based on vector SNCM-HMT and maximum likelihood (ML) decision rule. Experimental results on extensive test images demonstrate that the proposed statistical color image watermarking provides a performance better than that of most of the state-of-the-art statistical methods and some deep learning approaches recently proposed in the literature.

Attention enhanced machine instinctive vision with human-inspired saliency detection

Salient object detection (SOD) enables machines to recognize and accurately segment visually prominent regions in images. Despite recent advancements, existing approaches often lack progressive fusion of low and high-level features, effective multi-scale feature handling, and precise boundary detection. Moreover, the robustness of these models under varied lighting conditions remains a concern. To overcome these challenges, we present Attention Enhanced Machine Instinctive Vision framework for SOD. The proposed framework leverages the strategy of Multi-stage Feature Refinement with Optimal Attentions-Driven Framework (MFRNet). The multi-level features are extracted from six stages of the EfficientNet-B7 backbone. This provides effective feature fusions of low and high-level details across various scales at the later stage of the framework. We introduce the Spatial-optimized Feature Attention (SOFA) module, which refines spatial features from three initial-stage feature maps. The extracted multi-scale features from the backbone are passed from the convolution feature transformation and spatial attention mechanisms to refine the low-level information. The SOFA module concatenates and upsamples these refined features, producing a comprehensive spatial representation of various levels. Moreover, the proposed Context-Aware Channel Refinement (CACR) module integrates dilated convolutions with optimized dilation rates followed by channel attention to capture multi-scale contextual information from the mature three layers. Furthermore, our progressive feature fusion strategy combines high-level semantic information and low-level spatial details through multiple residual connections, ensuring robust feature representation and effective gradient backpropagation. To enhance robustness, we train our network with augmented data featuring low and high brightness adjustments, improving its ability to handle diverse lighting conditions. Extensive experiments on four benchmark datasets — ECSSD, HKU-IS, DUTS, and PASCAL-S — validate the proposed framework’s effectiveness, demonstrating superior performance compared to existing SOTA methods in the domain. Code, qualitative results, and trained weights will be available at the link: https://github.com/habib1402/MFRNet-SOD.