Soft Computing, Год журнала: 2023, Номер unknown
Опубликована: Май 5, 2023
Язык: Английский
Soft Computing, Год журнала: 2023, Номер unknown
Опубликована: Май 5, 2023
Язык: Английский
Signal Processing, Год журнала: 2025, Номер unknown, С. 110012 - 110012
Опубликована: Апрель 1, 2025
Язык: Английский
Процитировано
0Sensors, Год журнала: 2025, Номер 25(11), С. 3382 - 3382
Опубликована: Май 28, 2025
Numerous existing methods demonstrate impressive performance in brightening low-illumination images but fail detail enhancement and color correction. To tackle these challenges, this paper proposes a dual-branch network including three main parts: space transformation, correction (CC-Net), light-boosting (LB-Net). Specifically, we first transfer the input into CIELAB to extract luminosity components. Afterward, employ LB-Net effectively explore multiscale features via carefully designed large–small-scale structure, which can adaptively adjust brightness of images. And use CC-Net, U-shaped network, generate noise-free with vivid color. Additionally, an efficient feature interaction module is introduced for two branches’ information. Extensive experiments on low-light image public benchmarks that our method outperforms state-of-the-art restoring quality Furthermore, further indicate significantly enhances object detection under conditions.
Язык: Английский
Процитировано
0Applied Sciences, Год журнала: 2025, Номер 15(11), С. 6295 - 6295
Опубликована: Июнь 3, 2025
Infrared (IR) and visible (VIS) image fusion enhances vision tasks by combining complementary data. However, most existing methods assume normal lighting conditions thus perform poorly in low-light environments, where VIS images often lose critical texture details. To address this limitation, we propose VCAFusion, a novel approach for robust infrared scenarios. Our framework incorporates an adaptive brightness adjustment model based on light reflection theory to mitigate illumination-induced degradation nocturnal images. Additionally, design enhancement function inspired human visual perception recover weak further improve quality, develop edge-preserving multi-scale decomposition saliency-preserving strategy, ensuring seamless integration of perceptual features. By effectively balancing fusion, our preserves both the intensity distribution fine details salient objects. Extensive experiments public datasets demonstrate that VCAFusion achieves superior closely aligning with outperforming state-of-the-art qualitative quantitative evaluations.
Язык: Английский
Процитировано
0Infrared Physics & Technology, Год журнала: 2023, Номер 133, С. 104848 - 104848
Опубликована: Авг. 9, 2023
Язык: Английский
Процитировано
7IEEE Sensors Journal, Год журнала: 2023, Номер 23(3), С. 2783 - 2793
Опубликована: Янв. 4, 2023
The visible (VS) and near-infrared (NIR) image fusion is a common approach to improve visibility, which saves rich scene details similar colors the VS in fused image. However, preserving edge preventing color distortion fundamental yet challenging problem for VS-NIR work. To address this problem, article proposes novel method based on multiscale gradient guided edge-smoothing (MGES) model local weight. According spectrum characteristics of NIR, weight established by analyzing difference aiming only transfer prominent NIR image, thereby avoiding confusion spectral information. Furthermore, MGES designed simultaneously generate Laplacian pyramid domain filtering-based pyramid, fully considers correlation between neighboring pixels omits Gaussian filtering step, thus effectively spatial suppressing halo artifacts. Subjective objective experimental results demonstrate superiority proposed over state-of-the-art methods terms maintaining naturalness.
Язык: Английский
Процитировано
5Scientific Reports, Год журнала: 2024, Номер 14(1)
Опубликована: Янв. 2, 2024
Abstract The auto-encoder (AE) based image fusion models have achieved encouraging performance on infrared and visible fusion. However, the meaningful information loss in encoding stage simple unlearnable strategy are two significant challenges for such models. To address these issues, this paper proposes an model interactive residual attention contrastive learning frequency domain. Firstly, source is transformed into three sub-bands of high-frequency, low-frequency, mid-frequency powerful multiscale representation from prospective spectrum analysis. further cope with limitations straightforward strategy, a learnable coordinate module layer incorporated to adaptively fuse representative characteristics corresponding feature maps. Moreover, leveraged train decomposition network enhancing complementarity at different spectra. Finally, detail-preserving loss, jointly entire good detail maintainability. Qualitative quantitative comparisons demonstrate feasibility validity our model, which can consistently generate images containing both highlight targets legible details, outperforming state-of-the-art methods.
Язык: Английский
Процитировано
1Expert Systems with Applications, Год журнала: 2024, Номер 261, С. 125472 - 125472
Опубликована: Окт. 5, 2024
Язык: Английский
Процитировано
1Optics and Lasers in Engineering, Год журнала: 2023, Номер 170, С. 107765 - 107765
Опубликована: Июль 31, 2023
Язык: Английский
Процитировано
3Optics Express, Год журнала: 2023, Номер 32(2), С. 1650 - 1650
Опубликована: Дек. 20, 2023
In aerospace, the effects of thermal radiation severely affect imaging quality infrared (IR) detectors, which blur scene information. Existing methods can effectively remove intensity bias caused by effect, but they have limitations in ability enhancing contrast and correcting local dense or global intensity. To address limitations, we propose a enhancement method based on cyclic multi-scale illumination self-similarity gradient perception regularization solver (CMIS-GPR). First, conceive to correct for amplifying gradient. Specifically, (GPR) directly decomposing degraded image into pair high images, do not contain exhibit inverted directions. However, find that GPR fails area due small scene. Second, cope with cases intensity, regard as sum multiple slight bias. Then, construct (CMIS) model using Gaussian filters structural similarity prior removing layer layer. The result acts coarse correction GPR, does need be overly concerned whether has residuals not. Finally, corrected is input module further residual contrast. Extensive experiments real simulated data demonstrated superiority proposed method.
Язык: Английский
Процитировано
3Computers & Graphics, Год журнала: 2023, Номер 117, С. 124 - 133
Опубликована: Окт. 27, 2023
Язык: Английский
Процитировано
1