Retinex-based low-light image enhancement methods are widely used due to their excellent performance. However, most of them are time-consuming for large-size images. To solve this issue, this paper extends the Retinex model from the spatial domain to the histogram domain and proposes a novel histogram-based Retinex model for fast low-light image enhancement, named HistRetinex. First, we establish the relationship among the histograms of the illumination, reflectance, and low-light image based on the traditional Retinex model and an independence assumption, thereby approximating Retinex decomposition in the histogram domain. Second, based on prior information and the histogram-based Retinex model, we construct a novel two-level optimization model. By solving this optimization model, we derive the iterative formulas for the illumination and reflectance histograms, respectively. Finally, we enhance the low-light image by matching its histogram to the histogram estimated by HistRetinex. Experimental results demonstrate that the estimation errors between the estimated histogram and the low-light image histogram are within an acceptable range. Moreover, HistRetinex outperforms existing unsupervised enhancement methods in both visibility and performance metrics, while requiring only 1.85 seconds to process a $1000\times 664$ image, saving at least 6.68 seconds compared with the fastest competing traditional method. To further support efficient implementation, we use the MATLAB MEX tool to accelerate the implementation of HistRetinex, and the accelerated version runs at 28 FPS for $600\times 400$ images. The source code and experimental results are available at https://github.com/jingtianzhao/HistRetinex.

Intracerebral hemorrhage (ICH) necessitates precise and efficient segmentation of hemorrhagic regions in head computed tomography (CT) scans to facilitate timely clinical decisions. To address challenges such as irregular shapes of hematomas, unclear lesion boundaries, and the scarcity of annotated data, we introduce the ICH-PFNet, a text-guided segmentation framework specifically designed for ICH imaging that operates without prompts. The Mamba Pyramid Downsampling module ensures robust multi-scale feature extraction, while the GCS-CLIP fusion mechanism enhances semantic consistency through batch-level contrastive similarity. The Enhanced SAM module provides automatic spatial guidance and convolution-based sparse embeddings to eliminate manual input. Furthermore, a Feature Pyramid Network combined with a Group Aggregation Bridge enhances multi-scale feature fusion and refines boundaries. Our model showed superior performance in segmenting small and structurally complex hemorrhages by using a private CT dataset. These results highlight its potential for integration into automated ICH assessment workflows. The code is available at https://github.com/Hzchzc123/ICH-CMNet.

Emotion recognition based on EEG has attracted widespread research interest in the field of brain-computer interfaces. To extract EEG intra- and inter-channel features and find discriminative representations for EEG emotion recognition, we propose EEG-Multilayer Perceptron (EEG-MLP) architecture. EEG-MLP is completely composed of MLPs and mainly consists of two modules, one is a temporal mixer that captures intra-channel (temporal) information, and the other is a channel mixer that captures inter-channel information. The two modules learn knowledge in a parallel manner, and then their outputs are fused to extract global information and classify EEG emotions. We conduct extensive experiments on DEAP dataset. EEG-MLP is first compared with five inter-channel interaction models (related to CNN or GCN) to verify its effectiveness. Then, five other models with similar architecture to EEG-MLP were also contrasted. Experimental results show that EEG-MLP achieves the best performance among the above methods, with accuracies of 94.87% and 95.32% in the valence and arousal dimensions, respectively. In addition, it has a strong discrimination ability for complex categories, and has low requirements for storage resources.

Emotion recognition is a trending topic for research in the area of the brain computer interface (BCI). As an effective signal source, EEG(Electroencephalogram) is widely used in emotion recognition tasks, from which multiple features can be extracted in different domains, such as time domain and frequency domain. However, how to make full use of multiple domain features has become a challenge. In this study, we propose a transformer network for emotion recognition based on Multi-domain EEG features, named MEEG-Transformer. MEEG-Transformer can effectively capture the spatial information with the convolution layer, mine unique information within each domain, and explore the complementary information between features from different domains using self-attention mechanism. Specifically, we extract the features of time domain, frequency domain and wavelet domain respectively, construct the two-dimensional feature matrix of three domains based on the 10-20 system, and merge the three matrices into multi-domain EEG features. Using the DEAP dataset to perform experiments, the proposed model achieves 96.8% and 96.0% recognition accuracy in the arousal and valence dimensions respectively. It is indicated that the proposed method has a strong inspiration for emotion recognition tasks.

Research on human emotions based on EEG during multimedia stimuli is an emerging field that has made significant progress in EEG-based emotion classification. However, current studies often neglect the extraction of dynamic information from EEG signals and lack the exploration of local information. Moreover, many existing models are overly complex, demanding an excessive investment in training resources and time. In this paper, we propose a novel, simple two-stream channel reconstruction and feature attention network, named CRFAE-motionNet, for EEG emotion recognition. The main advantage of CRFAEmotionNet is its ability to simultaneously integrate static and dynamic information from EEG signals within a unified network. Additionally, it can extract continuous EEG temporal information through channel reconstruction and utilize feature attention to further explore local information. The proposed network was evaluated using the publicly available DEAP dataset. The experimental results indicate that the proposed CRFAEmotionNet outperforms the state-of-the-art baselines, achieving the accuracy of 98.7% for valence and 98.6% for arousal.

Empowering computer systems to automatically recognize human emotions has become an urgent need in the field of human-computer interaction (HCI). Two-dimensional emotion (Valence-Arousal) models are commonly used to represent emotions. Up to now, the correlation between emotion dimensions has rarely been investigated, and subject-independent EEG emotion recognition is still a challenging task. For this purpose, we introduce multi-task learning (MTL) into EEG emotion recognition. MTL learns different emotion dimensions simultaneously and extracts correlation information between dimensions in task-sharing space to coordinate the optimization of multiple emotion dimensions. We further propose Identity based Multi-gate Mixture-of-Experts (IDMMOE), which allocates part of model subspace for each subject in a customized manner according to the subject’s identity. Extensive experiments were conducted on DEAP dataset. Three MTL models were implemented: Shared-Bottom, Multi-gate Mixture-of-Experts, and Customized Gate Control respectively. They were compared with a single-task learning model trained separately on valence and arousal. Experimental results demonstrate that two emotion dimensions are intrinsically related, and MTL acquires such correlation information and improves prediction accuracy in both emotion dimensions. In addition, IDMMOE achieves average accuracies of 89.5% and 89.7% for valence and arousal respectively and it is effective for subject-independent experiment.

Spatiotemporal fusion is a topical framework for solving the mutual restricted problem between the spatial and temporal resolution of satellite images. We pioneer an approach to replace similarity measurement steps in spatiotemporal fusion algorithms with convolutional neural networks (CNNs), building a bridge between weight function-based models and the learning-based models. Specifically, we propose a nonlocal form that separates the relational computation part from the value representation part, and construct the CNN-based similarity weight learning block for learning normalized weights. The block can be inserted into spatial and temporal adaptive reflectance fusion model (STARFM) to replace the manually designed weight calculation rules common in weight function-based methods, or into the CNN model StfNet to better utilize neighboring high-resolution images. The trained model outputs a high-resolution prediction from each base date image pair. The final result is a combination of the two predictions. In this regard, we propose the standard deviation-based weights to combine two prediction results. Four experiments are performed on Landsat–Moderate-resolution Imaging Spectroradiometer (MODIS) image pairs to determine the following: 1) the performance of the model at the target training date; 2) the generalization of the model in the target training time period; and 3) the generalization of the model at different dates and different geographical locations, each considering the different cases of giving one and two pairs of known images. Experimental results demonstrate the superiority of the similarity weight learning block and standard deviation-based weights. Among them, STARFM with the similarity weight learning block exhibits strong generalization, which testifies to the practical value of our model.