Embedding watermarks in the diffusion latent space improves robustness but often alters visual content due to the distribution shift between watermarked and clean latent variables. To address this issue, stable latent integration watermark (SLIM) is proposed in this paper, in which watermarks are integrated into the features output by the noise prediction network of a diffusion model, while ensuring that the perturbation introduced in the diffusion latent space remains negligible. Specifically, a watermark encoder–decoder is first trained to convert binary watermark sequences into watermark latent variables that are dimensionally compatible with the diffusion latent variables, enabling flexible and reliable embedding and extraction. The watermark latent variables are processed through the first down-sampling block of the denoising U-Net, and the resulting watermark features are fused with the block output to minimize interference with image semantics. To counteract the perturbations in diffusion features induced by watermark embedding and to ensure accurate watermark extraction, the denoising U-Net is efficiently fine-tuned using a low-rank adaptation module. Experimental results demonstrate that SLIM achieves superior generation quality while exhibiting exceptional robustness against diverse attacks compared with baseline methods. Code will be available at https://github.com/XiaoxiKong/SLIM.

Novel view synthesis for underwater scene reconstruction presents unique challenges due to complex light-media interactions. Optical scattering and absorption in water body bring inhomogeneous medium attenuation interference that disrupts conventional volume rendering assumptions of uniform propagation medium. While 3D Gaussian Splatting (3DGS) offers real-time rendering capabilities, it struggles with underwater inhomogeneous environments where scattering media introduces artifacts and inconsistent appearance. In this study, we propose a physics-based framework that disentangles object appearance from water medium effects through tailored Gaussian modeling. Our approach introduces appearance embeddings, which are explicit medium representations for backscatter and attenuation, enhancing scene consistency. In addition, we propose a depth-guided optimization strategy that leverages pseudo-depth maps as supervision with depth regularization and scale penalty terms to improve geometric fidelity. By integrating the proposed appearance and medium modeling components via an underwater imaging model, our approach achieves both high-quality novel view synthesis and physically accurate scene restoration. Experiments demonstrate our significant improvements in rendering quality and restoration accuracy over existing methods. The project page is available at https://bilityniu.github.io/3D-UIR.

Underwater exploration is crucial for geoscience and remote sensing, but the capture of underwater images is compromised by the degradation of light absorption and scattering. This article proposes a diffusion-color-guided framework (DCGF) to enhance the quality of underwater images and address color deviations caused by randomness in general diffusion models during underwater image restoration. In DCGF, the diffusion model reconstructs the image distribution, while a color correction module ensures accurate color representation. A conditional image guides the denoising procedure, aligning the diffusion trajectory closely with the target domain. This approach reduces the impact of diffusion variability and minimizes deviations. Once a predetermined denoising threshold is reached, the color correction module extracts salient characteristics of color distribution from luminance and RGB channels, enhancing overall efficacy. The experimental results demonstrate that the DCGF algorithm effectively restores degraded underwater images with robustness and effectiveness. The method successfully corrects color degradation and recovers details in low-light conditions, significantly improving underwater image quality.

Underwater images often suffer from significant information loss in the red color channel, resulting in a predominantly bluish or greenish tone. Existing enhancement methods struggle to address this issue due to uniform enhancement applied to the bluish and greenish channels, resulting in overcompensation or under-compensation in the red channel. To address these challenges and achieve a more natural color restoration in underwater images, we propose the adaptive color compensation and enhancement (ACCE) algorithm. The ACCE algorithm comprises several essential steps. Initially, to recover the loss of red channel information more effectively, we divide the images into bluish and greenish components for preliminary color compensation (PCC) in the RGB color space. Subsequently, we introduce a novel minimum color loss (MCL) constraint to regulate the PCC, ensuring balanced histogram distributions across the RGB channels. Furthermore, for improved color balance in the enhanced underwater image, we design the fine-tuning color compensation (FCC) to the$a$and$b$channels of the CIELAB color space. Ultimately, we employ the Contour Bougie (CB) enhancement algorithm to restore contour details in underwater images. Experimental results validate the superiority of the proposed ACCE algorithm over state-of-the-art methods, as demonstrated through qualitative and quantitative comparisons. In addition, ACCE exhibits promising generalization and potential for broader applications, encompassing tasks such as dehazing and lowlight image enhancement.

Spaceborne infrared ship detection holds immense research significance in both military and civilian domains. Nonetheless, the focus of research in this field remains primarily on optical and synthetic aperture radar (SAR) images due to the confidentiality and limited accessibility of infrared data. The challenges in spaceborne ship detection arise from the long-distance capture and low signal-to-noise ratio of infrared images, which contribute to false alarm misclassifications. To handle this problem, this article concentrates on enhancing information interaction during feature extraction to discern disparities between targets and backgrounds more effectively, and we propose a multitype feature perception and refined network (MFPRN). Specifically, we propose a dual feature fusion scheme, which combines a fast Fourier (FF) module used to obtain comprehensive receptive field and a lightweight Multilayer Perceptron (MLP) applied to capture the long-range feature dependencies. Besides, we adopt a Cascade region proposal network (RPN) to leverage high-quality region proposals for the prediction head. Through the extraction of rich features and refined candidate boxes, we successfully mitigate false alarms. Experimental results illustrate that our method significantly reduces false alarms for general detectors, culminating in state-of-the-art performance as demonstrated on the public infrared ship detection dataset (ISDD) baseline.

Underwater detection is a crucial task to lay the foundation for the intelligent marine industry. In contrast to land scenes, targets in degraded underwater environments show ambiguous and surrounding-similar profiles, causing it challenging for generic detectors to accurately extract features. Eliminating the interference of ambiguous features is one of the primary goals when recognizing underwater objects against complex backgrounds. To this aim, we propose a novel detection framework called underwater distraction mining detector (UDMDet). UDMDet is an end-to-end detector and has two key modules: distraction-aware FPN (DAFPN) and task-aligned head (THead). DAFPN is designed to progressively refine the coarse features via mining the discrepancies between objects and backgrounds, while THead enhances the information interaction between classification and localization to make predictions with higher quality. To overcome the feature ambiguous problem, the underwater distraction-aware model is proposed to extract the differences between objects and surroundings so as to clear the target boundary. Experimental results show that UDMDet can more effectively discover objects conceal on real-world underwater images and has a higher precision outperforming the state-of-the-art detectors.

SAR ship detection is sensitive to the interference of inshore background, disturbance of strong wind and waves. The similar textures of the neighbor objects in SAR images affect the detection performance. As a remarkable indicator, textural information in the frequency domain characterizes the subtle textural differences between an object and its surroundings. Inspired by this, a multi-feature fusion network (MFFN) for SAR ship detection is constructed in this paper, which can obtain contour and detail information of a SAR image for detecting ships from their background. Firstly, spatial and frequency information of ship targets, which characterizes the whole and subtle textural information of ship targets, are extracted by a double-backbone network with Haar wavelet transform. Afterward, a binary domain feature pyramid network (BDFPN) with feature fusion block (FFB) is applied to fuse the spatial, frequency textural information of ship targets to obtain the fused feature maps with a top-down structure. Finally, those feature maps are adopted through the region proposal network for detecting ship targets from original images. The experimental results show that the proposed method achieves greater performance and more accurate detection results in unique situations in the state-of-the-art SAR ship detection data set (SSDD).

Light traveling through water results in strong scattering across color channels, restricting visibility in underwater images. Many cutting-edge underwater image enhancement methods encounter limitations in color recovery accuracy and resilience against irrelevant feature interference. To tackle these degradation challenges, we propose an efficient and fully guided information flow network called UGIF-Net, for enhancing underwater images. Specifically, we propose a multi-color space-guided color estimation module that accurately approximates color information by incorporating features from two color spaces within a unified network. Subsequently, we employ a dense attention block to guide the network in thoroughly extracting color information from both color spaces while adaptively perceiving crucial color information. Moreover, we devise a color-guided map to steer the network’s focus toward color information and augment its response to color quality degradation. We incorporate the guided map into a guide color restoration module to achieve visually appealing enhancement results. Comprehensive experiments indicate that our approach surpasses state-of-the-art methods, showcasing favorable image restoration effects and their potential to aid other high-level vision tasks.

Low contrast, color distortion, and blurred details are common problems that perplex vision-guided underwater robots. To this end, we propose an underwater image enhancement framework via self-attention and contrastive learning (UIESC) to solve these problems. In this article, local features and global dependencies are constructed through space and channel dual attention, and criss-cross attention is used to solve the high computational complexity of self-attention. Moreover, contrastive learning is introduced into network training as a loss function, and contrastive regularization ensures that the enhanced images are closer to clear positive samples and away from source negative samples. Finally, smoothed-histogram equalization is adopted for further optimization to accommodate complex and variable underwater scenes. Extensive experiments have shown that our framework outperforms state-of-the-art methods in underwater image enhancement tasks.

Real-world underwater images suffer from quality degeneration caused by the scattering and absorption of light propagation. The damage of the detailed textures in underwater images shows the negative effect of detection and recognition. To recovery the image visibility and sharpness for the above applications, a new image enhancement method is proposed for extracting the image textures. To enhance the image textures with high quality, we propose a multiscale fusion enhancement. Two new fusion inputs are built on different color methods. One input is devoted to improve the sharpness by contrast-based dark channel prior dehazing in the red–green–blue (RGB) model. The other input is designed based on multiple morphological operation and color compensation from the opponent color in the CIE$1976~L^{\ast}a^{\ast}b^{\ast}$color space (CIELAB) model. This input is used to enhance the counter brightness and adjust the color distribution. The dominant features of the two inputs are merged. Therefore, the contrast of the fusion output is enhanced adaptively to recover the final enhanced result. Compared with the state-of-the-art methods, our results reveal that the proposed method can enrich the image textures based on an impressive visual perception of contrast, saturation, and sharpness. Moreover, our method also shows strong robustness in challenge scenes and improves the performance of several underwater applications.

Underwater images require further enhancement to improve the image qualities caused by medium scattering and light absorption. Based on Contour Bougie (CB) morphology, we propose a new enhancement method to enhance the scene contours and improve the visibility of images captured underwater. Two structuring elements with different sizes are considered as the roving windows. Multiple morphological operations are designed for highlighting the rich details on the origin images. The enhanced images are normalized and stretched to improve the white balance of RGB channels. The comprehensive study of state-of-the-art algorithms is conducted to interpret the improvement of image quality by the proposed method. In addition, we use 890 raw underwater degraded images as the testing data. The quantitative and qualitative evaluations of these data demonstrate that the proposed method achieves better visible contrast for highlighting the details of the undersea creatures. The comparison with different underwater scenes proves that the proposed method improves the color balances of the degraded images.