Introduction We previously investigated color constancy in photorealistic virtual reality (VR) and developed a Deep Neural Network (DNN) that predicts reflectance from rendered images. Methods We combine both approaches to compare and study a model and human performance with respect to established color constancy mechanisms: local surround, maximum flux and spatial mean. Rather than evaluating the model against physical ground truth, model performance was assessed using the same achromatic object selection task employed in the human experiments. The model, a ResNet based U-Net from our previous work, was pre-trained on rendered images to predict surface reflectance. We then applied transfer learning, fine-tuning only the network's decoder on images from the baseline VR condition. To parallel the human experiment, the model's output was used to perform the same achromatic object selection task across all conditions. Results A strong correspondence between the model and human behavior was observed. Both achieved high constancy under baseline conditions and showed similar, condition-dependent performance declines when the local surround or spatial mean color cues were removed. Discussion These results show that a pixel-wise DNN trained on naturalistic image statistics can reproduce the structure of human color constancy behavior across controlled cue manipulations, supporting the view that human constancy can arises from the integration of multiple scene-based cues without explicit illuminant estimation.

Accurate estimation of reflectance and illumination maps in the Retinex framework remains a significant challenge for low-light image enhancement due to inherent decomposition ambiguity. To address this, we propose DualPrior-Retinex, a novel framework that, inspired by Retinex theory, leverages the practical advantages of the YUV color space for robust enhancement. Our framework introduces a dual-prior architecture that effectively decouples the restoration process. It combines a diffusion-based global prior, responsible for ensuring low-frequency content consistency, with a YUV-based local prior designed to preserve high-frequency structural details. These complementary components are integrated by our Hierarchical Prior Fusion Module (HPFM), which balances perceptual quality with pixel-level fidelity in complex low-light scenarios. Extensive evaluations on multiple benchmarks demonstrate that our method achieves state-of-the-art performance across diverse metrics and visual qualities. Codes will be released at https://github.com/I2-Multimedia-Lab/DP-Retinex.

Abstract Traditional photometric stereo techniques are constrained by stringent hardware calibration requirements, pronounced sensitivity to specular highlights and cast shadows, and an inherent difficulty in reconciling reconstruction fidelity with practical usability. These limitations significantly impede their deployment in high-precision 3D facial reconstruction. To address these challenges, this study presents an uncalibrated photometric stereo framework specifically designed for facial scenarios, integrating two key modules within a two-stage optimization pipeline. In the first stage, illumination directions are estimated using a shared-weight convolutional architecture with global max-pooling and self-attention mechanisms to extract illumination-invariant features. In the second stage, PCA-based whitening followed by Huber robust regression is employed to recover surface normals under uncertain illumination conditions. Comprehensive experiments on synthetic and real-world datasets demonstrate the effectiveness of the proposed approach. On the test set, the illumination estimation network achieves a mean angular error of 7.29° for lighting direction prediction, while intensity estimation errors are reduced by 25%–33% compared with existing baselines. For surface normal reconstruction, the proposed method achieves mean angular errors of 6.7° under the Oracle-Dir setting (using ground-truth lighting directions) and 7.6° under the Est-Dir setting (using estimated lighting directions). By eliminating the need for precise hardware calibration, the framework reduces system complexity and operational cost while improving robustness, making it well suited for applications such as 3D facial recognition and digital human modeling. Future work will explore unified end-to-end optimization strategies to further improve reconstruction accuracy and generalization capability.

Boosting illuminant estimation in deep color constancy through brightness robustness enhancement

Zero-reference illumination estimation model for image enhancement in underground mines

Removing scattering is a central challenge in underwater imaging because water induces complex light-matter interactions. Scattering not only blurs image details but also degrades the reliability of downstream vision tasks. Importantly, backscattered light and object-reflected light exhibit distinct polarimetric differences, offering a physical cue for descattering. However, most existing approaches usually rely on a single polarization parameter or assume a constant background polarization degree, limiting their effectiveness in complex scenes. Therefore, an underwater image enhancement network based on a diffusion model and a polarization drive is proposed, which decomposes the problem into two stages: (i)descattering prior generation and (ii)scattering removal. In the prior stage, the diffusion model is introduced for the first time, to our knowledge, to fully mine the polarization information and distill it into high-quality priors that guide accurate scattering suppression. In the removal stage, the illumination estimation module is designed to compensate for water-body absorption and enforce global illumination consistency, thereby improving perceptual naturalness. Extensive experiments demonstrate state-of-the-art quantitative and qualitative performance, providing reliable, high-quality inputs for downstream underwater vision applications.

Computer vision and deep learning are essential in diverse fields such as autonomous driving, medical imaging, face recognition, and object detection. However, enhancing low-light remote sensing images remains challenging for both research and real-world applications. Low illumination degrades image quality due to sensor limitations and environmental factors, weakening visual fidelity and reducing performance in vision tasks. Common issues such as insufficient lighting, backlighting, and limited exposure create low contrast, heavy shadows, and poor visibility, particularly at night. We propose QWR-Dec-Net, a quaternion-based Retinex decomposition network tailored for low-light image enhancement. QWR-Dec-Net consists of two key modules: a decomposition module that separates illumination and reflectance, and a denoising module that fuses a quaternion holistic color representation with wavelet multi-frequency information. This structure jointly improves color constancy and noise suppression. Experiments on low-light remote sensing datasets (LSCIDMR and UCMerced) show that QWR-Dec-Net outperforms current methods in PSNR, SSIM, LPIPS, and classification accuracy. The model’s accurate illumination estimation and stable reflectance make it well-suited for remote sensing tasks such as object detection, video surveillance, precision agriculture, and autonomous navigation.

ABSTRACT Illuminant estimation in camera color space is a critical step in the image signal processing pipeline, which is commonly performed based on the RGB RAW images captured by a camera. Recently, most smartphones are equipped with an ambient light sensor (ALS) around the camera. In this work, we propose an illuminant estimation method (i.e., RACC) using both RGB RAW images and ALS signals as the inputs. A large dataset containing both RGB RAW images and ALS signals of more than 1300 scenes was collected, which was used to develop and evaluate the proposed RACC method. The results show that the RACC method results in much smaller angular errors and requires much fewer computational resources. In particular, the method was designed with a training mechanism to consider a practical challenge of incomplete or missing ALS signals under low light conditions, which was verified to have stable performance. These clearly suggest that the RACC method can be deployed for practical applications.

This paper presents a Decoupling-Centric Decomposition network for Low-Light Image Enhancement (DCD-Net). The DCDNet addresses two key challenges: (1) existing methods center on how to design the enhancement network and ignore the decomposition network’s critical role to decouple reflectance and illuminations as the first step and (2) existing decomposition networks process images directly (or with pre-denoising), ignore their compliance with the Retinex theory. Specifically, by centering on illumination–reflectance decoupling refinement, DCDNet operates without reliance on supplementary enhancement networks. It consists of a preprocessing network and a decomposition network. The preprocessing network adopts a self-supervised learning mechanism to suppress Retinex-incompatible features in the input image, thereby improving the quality of Retinex decomposition. Within the decomposition network, the reflectance net is designed to suppress the contamination of illumination on reflectance restoration by the Dual-Gated Directional Reflectance Module (DGD-RM) and Reflectance-Guided Multi-head Self-Attention (RG-MSA), while the illumination net utilizes DCT to achieve local–global illumination estimates. Comprehensive experiments were conducted on two benchmark datasets (LOL and MIT) and five unpaired datasets. The quantitative results on different datasets are as follows (measured by PSNR and SSIM): LOL v1 (20.87, 0.770) and MIT (21.66, 0.864). The average NIQE across five unpaired datasets is 3.420. Both qualitative and quantitative analyses demonstrated the superiority of DCDNet over state-of-the-art methods. Moreover, ablation studies demonstrated the effectiveness of each module in DCDNet.

Vessel identification in low-light environments is a challenging task since low-light images contain less information for detecting objects. To improve the feasibility of vessel identification in low-light environments, we present a new unsupervised low-light image augmentation approach to augment the visibility of vessel features in low-light images, laying a foundation for subsequent identification. This guarantees the feasibility of vessel identification with the augmented image. To this end, we design an illumination estimation network (IEN) to estimate the illumination of a low-light image based on the Retinex theory. Then, we augment the low-light image by estimating its reflectance with the estimated illumination. Compared with the existing deep learning-based supervised low-light image augmentation approach that depends on the low- and normal-light image pairs for model training, IEN is an unsupervised approach without using normal-light image as references during model training. Compared with the traditional unsupervised low-light image augmentation approach, IEN shows faster image augmentation speed by parallel computation acceleration with image Processing Units (GPUs). The proposed approach builds an end-to-end pipeline integrating a vessel-aware weight matrix and SmoothNet, which optimizes illumination estimation under the Retinex framework. To evaluate the effectiveness of the proposed approach, we build a low-light vessel image set based on the Sea Vessels 7000 dataset—a public maritime image set containing 7000 vessel images across multiple categories Then, we carry out an experiment to evaluate the feasibility of vessel identification using the augmented image. Experimental results show that the proposed approach boosts the AP75 metric of the RetinaNet detector by 6.6 percentage points (from 56.8 to 63.4) on the low-light Sea Vessels 7000 dataset, confirming that the augmented image significantly improves vessel identification accuracy in low-light scenarios.

We discovered a new type of assimilative color induction. An achromatic target with a white background was placed in the center of a concentric chromatic gradient that caused the glare effect. The target frequently appears to be in the same hue as the gradient. We discussed lower-level factors such as lateral inhibition and spatial summation functions, and higher-level factors such as illumination estimation.

Eliminating the color cast of the illuminant is a critical step in modern image processing systems, which has been addressed with a great number of illuminant estimation algorithms. The algorithms are found not effective for some specific contexts and applications, which leads to the development of scenarios-specific algorithms leveraging domain-specific cues. This paper investigated how facial cues help illuminant estimation. A total of 1299 images were captured under various dual-illuminant conditions, including real-world environments and lab settings. Modifications were made on existing methods by considering the facial information, which resulted in better performance.

Accurate white balancing (WB) remains a critical challenge in image signal processing. Particularly white point estimation for scenes with limited or ambiguous color information can lead to color casts and degraded visual quality. Moreover, many every-day scenes contain multiple relevant illuminants, further complicating illuminant estimation. Estimating multiple white points in a scene exacerbates the challenge and traditional RGB-based WB algorithms struggle with the limited information available for localized regions of a scene. We introduce a novel approach leveraging compact multi-spectral camera technology to improve local WB performance. By capturing additional, narrow-band spectral information beyond the RGB channels, our method provides more accurate white point estimation. We present comparative results demonstrating the advantages of multi-spectral sensing over conventional approaches, highlighting its potential to enable more intelligent and adaptive imaging pipelines in mobile, automotive, and industrial applications. Our approach is based on a relatively simple neural network, trained on simulated multi-spectral measurements. We developed a data collection protocol to establish a medium-sized validation dataset for which we report white point angles and color accuracy (DE2000) values. Our study includes a performance benchmark against a state-of-the art deep learning-based algorithm for our new validation set and the publicly available Large Scale Multi-Illuminant (LSMI) dataset. From the results we observe that our network performs on par with the state-of-the-art algorithm on the LSMI dataset and outperforms the state-of-the-art algorithm on our validation dataset. Furthermore, the multi-spectral based network outperforms the RGB based network.

In the context of visual effects and computer graphics, image generation and image enhancement are one of the basic problems, Same situation in other areas like space science and medicinal science. The picture quality is impacted by the presence of light in the image. This study introduces a unique method for generating and enhancing low-illumination photographs and generation of brighter time image it implies the usage of a deep mastering set of rules, particularly a deep convolutional Wasserstein generative adversarial network (DC-WGAN). The method involves changing snap shots from RGB to CIELAB shade space, which aligns extra intently with human visible belief, taking into account specific illumination estimation and mitigation of uneven lighting fixtures outcomes. By employing DC-WGAN to decorate the brightness aspect via an extended era community, the algorithm captures and amplifies important photograph features. The stronger LAB photographs are then converted back to RGB area to produce the final improved photos. The effectiveness of this approach is tested thru experiments under well known, special, and realistic conditions, demonstrating advanced overall performance as compared to 4 typically used algorithms. This study affords an important technological advancement for enhancing robot target reputation and renovation operations in area environments.

Convolutional Neural Networks (CNNs) have shown significant success in the low-light image enhancement task. However, most of existing works encounter challenges in balancing quality and efficiency simultaneously. This limitation hinders practical applicability in real-world scenarios and downstream vision tasks. To overcome these obstacles, we propose a Self-Calibrated Illumination (SCI) learning scheme, introducing a new perspective to boost the model's capability. Based on a weight-sharing illumination estimation process, we construct an embedded self-calibrator to accelerate stage-level convergence, yielding gains that utilize only a single basic block for inference, which drastically diminishes computation cost. Additionally, by introducing the additivity condition on the basic block, we acquire a reinforced version dubbed SCI++, which disentangles the relationship between the self-calibrator and illumination estimator, providing a more interpretable and effective learning paradigm with faster convergence and better stability. We assess the proposed enhancers on standard benchmarks and in-the-wild datasets, confirming that they can restore clean images from diverse scenes with higher quality and efficiency. The verification on different levels of low-light vision tasks shows our applicability against other methods.

Abstract This paper addresses the challenge of accurately estimating both the full illuminant spectral power distribution (SPD) and the per-pixel spectral reflectance from an RGB image captured with a known camera. Full spectral information allows us to perform more accurate white balance, or render the scene under another illuminant. Traditional color constancy methods focus on predicting the illuminant color as a 3-dimensional projection of the infinite-dimensional illuminant SPD onto the camera spectral sensitivity functions (SSFs) space. Because different illuminants can have the same projection in the 3-dimensional SSFs space, those traditional methods cannot differentiate between such illuminants (metamers in the camera response space) and hence remove the color cast resulting from the illumination. We reconstruct the spectral information using a neural network with two interconnected branches: one branch predicts the illuminant SPD, and the other reconstructs the per-pixel spectral reflectance by integrating the predicted SPD within its intermediate layers. Experimental results demonstrate that our framework achieves superior performance in estimating both the illuminant SPD and the per-pixel spectral reflectance compared to the previous approach.

Adaptive structural compensation enhancement based on multi-scale illumination estimation

Abstract Following the tradition of previous year’s successful research events dedicated to the future of imaging science, the 6th London Imaging Meeting (LIM) takes place at the Institute of Physics (IoP) in London, UK, from September 8 to 10, 2025. The theme chosen for this year is “Lighting and Imaging” with the intention of bringing together students, researchers, engineers and others from these two distinct fields of science and technology, which share so many fundamental concepts and methods, and whose research challenges are often deeply intertwined. Just think about trying to capture good quality images in really low light levels! The first day features a summer school with four insightful short courses on radiometry, photometry and colour science for lighting; materials, lights, algorithms, and software for light transport simulation; the colourful math behind multichannel imaging systems; and methods for illuminant estimation and correction in colour imaging workflows. List of Conference Chairs, Series chairs and Steering Committee are available in this PDF.

Variations in illumination conditions critically degrade color fidelity in digital images, thereby compromising the accuracy of downstream computer vision tasks. Building upon these historical principles, this paper proposes a self-attention autoencoding feature support vector regression algorithm. The method extracts probability distributions in the luminance-red-green color space as primitive features, reconstructs them through a self-attention augmented autoencoder, and deploys support vector regression for illumination estimation. Experimental validation demonstrates superior robustness against noise and illumination diversity compared to feature-based alternatives. On the linear GreyBall SFU dataset, our method achieves an average 64.4% reduction across six key error metrics relative to the minimum values from all eight comparative methods. On the Cube++ dataset, it yields an average 44.9% reduction across six error metrics relative to the minimum values from all six comparative methods.

Imaging technologies are widely used in surveillance, medical diagnostics, and other critical applications. However, under low-light conditions, captured images often suffer from insufficient brightness, blurred details, and excessive noise, degrading quality and hindering downstream tasks. Conventional low-light image enhancement (LLIE) methods not only require annotated data but also often involve heavy models with high computational costs, making them unsuitable for real-time processing. To tackle these challenges, a lightweight and unsupervised LLIE method utilizing a dual-stage frequency-domain calibration network (DFCNet) is proposed. In the first stage, the input image undergoes the preliminary feature modulation (PFM) module to guide the illumination estimation (IE) module in generating a more accurate illumination map. The final enhanced image is obtained by dividing the input by the estimated illumination map. The second stage is used only during training. It applies a frequency-domain residual calibration (FRC) module to the first-stage output, generating a calibration term that is added to the original input to darken dark regions and brighten bright areas. This updated input is then fed back to the PFM and IE modules for parameter optimization. Extensive experiments on benchmark datasets demonstrate that DFCNet achieves superior performance across multiple image quality metrics while delivering visually clearer and more natural results.