ABSTRACT Colour differences often exist between remote sensing images. Moreover, images of Antarctic region also suffer from uneven luminance due to its high albedo. In this letter, an illumination normalization method for Antarctic remote sensing images is proposed, using a binary constrained Gaussian filter (BCGF) for within-image luminance normalization and a colour constancy method for between-image colour normalization. BCGF is the product of a Gaussian filter and the binary luminance map of the image. Testing on HY-1C/D satellite images of Antarctica demonstrates the effectiveness of the proposed method in terms of illumination normalization, texture preservation, and gradient preservation, outperforming several existing methods.

Color imaging algorithms - such as color correction, spectral estimation and color constancy - are developed and validated with spectral reflectance data. However, the choice of the reflectance data set - used in development and tuning - not only affects the results of these algorithms but it also changes the ranking of the different approaches. We propose that this fragility is because it is difficult to measure/sample enough data to statistically represent the large number of degrees of freedom apparent in spectral reflectances. In this paper, we propose that the space of reflectance data should not be sampled but, rather, integrated. Specifically, we advocate that the convex closure of a reflectance data set - all convex combinations of all spectra - should be used instead of discrete reflectance samples. To make the integration computation tractable, we approximate these convex closures by their enclosing hyper-cube in a privileged coordinate system. We use color correction as an exemplar color imaging problem to demonstrate the utility of our approach.

Color constancy, the human visual system's ability to perceive consistent colors under varying illumination conditions, is crucial for accurate color perception. Recently, deep learning algorithms have been introduced into this task and have achieved remarkable achievements. However, existing methods are limited by the scale of current multi-illumination datasets and model size, hindering their ability to learn discriminative features effectively and their practical value for deployment in cameras. To overcome these limitations, this paper proposes a multi-illumination color constancy approach based on self-supervised learning and knowledge distillation. This approach includes three phases: self-supervised pre-training, supervised fine-tuning, and knowledge distillation. During the pre-training phase, we train Transformer-based and U-Net based encoders by two pretext tasks: light normalization task to learn lighting color contextual representation and grayscale colorization task to acquire objects' inherent color information. For the downstream color constancy task, we fine-tune the encoders and design a lightweight decoder to obtain better illumination distributions with fewer parameters. During the knowledge distillation phase, we introduce a hybrid knowledge distillation technique to align CNN features with those of Transformer and U-Net respectively. Our proposed method outperforms state-of-the-art techniques on multi-illumination and single-illumination benchmarks. Extensive ablation studies and visualizations confirm the effectiveness of our model.

Color constancy seeks to keep the perceived color of objects consistent under varying illumination conditions. However, existing methods often rely on restrictive prior assumptions or suffer from limited generalization capability, posing significant challenges in complex scenes with multiple light sources. In this paper, we propose a neural network-enhanced, physicsbased approach to multi-illuminant color constancy that leverages spectral imaging—highly sensitive to illumination variation. First, we analyze the physical image-formation process under mixed lighting and introduce a master–subordinate illumination model, extending conventional correlated-color-temperature reillumination techniques. Our neural network framework ex-plicitly models the correlation between narrow-band spectral reflectance and the spectral power distribution (SPD) of the illumination, enabling accurate recovery of the scene light's full SPD. Using this model, we fuse RGB images with the estimated illumination spectra to predict illuminant chromaticity precisely, then correct image colors to a standard reference light. Extensive experiments on synthetic multi–color-temperature datasets and real-world spectral datasets demonstrate that our neural network-based method achieves state-of-the-art accuracy in spectral estimation and color-constancy correction.

This paper presents an empirical investigation into illuminant estimation using multi-spectral images. Our study emphasizes two key contributions: (1) the utilization of the estimated multi-spectral images and (2) the incorporation of a hierarchical structure. Firstly, exploiting multi-spectral images proves to have a positive influence on illuminant estimation, particularly in scenarios characterized by monochromatic images where conventional color constancy methods face challenges. Our experimental results vividly illustrate the effectiveness of leveraging spectral information in enhancing illuminant estimation. Secondly, the adoption of a hierarchical structure stems from the need for spatial invariance in the task of estimating a global illuminant. To further enhance the performance of the hierarchical structure, we employ a contrastive loss applied to different scaled outputs. This approach demonstrates remarkable effectiveness on our custom dataset, showcasing superior performance compared to the existing methods. In addition, we extend the evaluation to the widely recognized NUS-8 dataset, where the proposed method showcases a notable 26.7% relative improvement over the previous state-of-the-art methods.

ABSTRACTSkin color constancy under nonuniform correlated color temperatures (CCT) and multiple light sources has always been a hot issue in color science. A more high‐quality skin color reproduction method has broad application prospects in camera photography, face recognition, and other fields. The processing process from the 14bit or 16bit RAW pictures taken by the camera to the final output of 8bit JPG pictures is called the image processing pipeline, in which the steps of the auto‐white balance algorithm have a decisive impact on the skin color reproduction result. The traditional automatic white balance algorithm is based on hypothetical statistics. Moreover, the estimated illuminant color is obtained through illuminant estimation. However, the traditional grayscale world, perfect reflector, and other auto‐white balance algorithms perform unsatisfactorily under non‐uniform or complex light sources. The method based on sample statistics proposes a new solution to this problem from another aspect. The deep learning algorithm, especially the generative adversarial network (GAN) algorithm, is very suitable for establishing the mapping between pictures and has an excellent performance in the fields of image reconstruction, image translation, defogging, and coloring. This paper proposes a new solution to this problem. The asymmetric UNet3+ shape generator integrates better global and local information to obtain a more refined correction matrix incorporating details of the whole image. The discriminator is Patch‐discriminator, which focuses more on image details by changing the attention field. The dataset used in this article is the Liverpool‐Leeds Skin‐color Database (LLSD) and some supplementary images, including the skin color of more than 960 subjects under D65 and different light sources. Finally, we calculate the CIEDE2000 color difference and some other image quality index between the test skin color JPEG picture corrected by the auto‐white balance algorithm and the skin color under the corresponding D65 to evaluate the effect of white balance correction. The results show that the asymmetric GAN algorithm proposed in this paper can bring higher quality skin color reproduction results than the traditional auto‐white balance algorithm and existing deep learning WB algorithm.

Abstract The accurate detection of moving objects is essential in various applications of artificial intelligence, particularly in the field of intelligent surveillance systems. However, the moving cast shadow detection significantly decreases the precision of moving object detection because they share similar motion characteristics. To address the issue, the authors propose an innovative approach to detect moving cast shadows by combining the hybrid feature with a broad learning system (BLS). The approach involves extracting low‐level features from the input and background images based on colour constancy and texture consistency principles that are shown to be highly effective in moving cast shadow detection. The authors then utilise the BLS to create a hybrid feature and BLS uses the extracted low‐level features as input instead of the original data. BLS is an innovative form of deep learning that can map input to feature nodes and further enhance them by enhancement nodes, resulting in more compact features for classification. Finally, the authors develop an efficient and straightforward post‐processing technique to improve the accuracy of moving object detection. To evaluate the effectiveness and generalisation ability, the authors conduct extensive experiments on public ATON‐CVRR and CDnet datasets to verify the superior performance of our method by comparing with representative approaches.

Improving low-light images to enhance prediction in various applications has a greater advantage. A two-pronged approach that employs Deep Convolutional Generative Adversarial Networks with weight regularization (DCGAN-WR) and Zero-Reference Deep Curve Estimation (DCE) was used. The model was trained using the LOL dataset, and the results showed significant improvements in image quality. The DCGAN is fine-tuned with Group Lasso regularization to enhance the performance. The DCGAN-WR model is shown to enhance images realistically, demonstrating its capacity to learn characteristics and texture representations from low-light input. Empirical and simulated image comparisons demonstrate remarkable performance under demanding low-light settings. Moreover, the DCE model employs a novel approach that considers color constancy loss, light smoothness, and spatial consistency. Information about the learning dynamics and curve parameter changing capabilities of the model can be visualized by loss function graphs, which aim to maximize picture quality. Compared to the original images, Images generated by the DCE models maintain color accuracy, increase exposure levels, and preserve spatial coherence. A solution for low-illumination image enhancement is achieved through the proposed model DCGAN-WR and DCE. Genuine details are recorded by the GAN model, while the DCE adjusts the exposure levels and color balance to produce improved, aesthetically pleasing, and contextually accurate images. The proposed approach not only outperforms the other methods on the LOL dataset but also exhibits potential for practical use in computer vision tasks, which require higher image quality for precise analysis and interpretation.

Unharvested palm fruit bunch (PFB) ripeness detection is crucial for efficient palm oil production as labor costs rise and worker shortages grow. Challenges remain under varying illumination, affecting detection performance. Although the deep learning-based color constancy model shows promise, the limitation lies in the lack of data. This paper addresses these issues by proposing a hybrid color correction method that combines the learning-based and the physical-based color constancy technique, then trains with YOLOv8 to enhance unharvested PFB detection performance. Specifically, the captured images and their corresponding simultaneous spectra are used to generate the ground truth images, then map the images to other ambient spectra to train the color constancy model, and finally, train the YOLOv8 with the corrected images for ripeness detection. Results demonstrate a 1.5% improvement in the mean Average Precision (mAP) @0.5 of YOLOv8, increasing from 88.2% to 89.7% after applying the proposed hybrid color correction method. For each ripeness category, the mAP@0.5 for unripe and ripe bunches increased by 1.7% and 1.8%, respectively. Underripe bunches exhibit the most increment in mAP of 2.1%, while the overripe bunch gives the least increase of 0.4%. The qualitative results show that the model has enhanced the ability to distinguish subtle differences in PFB ripeness. Consequently, it advances the effectiveness and reliability of PFB ripeness detection in practical palm oil production scenarios, especially for unripe to ripe bunches wherein color gives the most information. This research demonstrates potential in object detection tasks that incorporate illumination-based color correction methods.

An application oriented all-round intelligent weeding machine with enhanced YOLOv5

The proliferation of AI-generated content (AIGC) has empowered non-experts to create highly realistic Deepfake images and videos using user-friendly software, posing significant challenges to the legal system, particularly in criminal investigations, court proceedings, and accident analyses. The absence of reliable Deepfake verification methods threatens the integrity of legal processes. In response, researchers have explored deep forgery detection, proposing various forensic techniques. However, the swift evolution of deep forgery creation and the limited generalizability of current detection methods impede practical application. We introduce a new deep forgery detection method that utilizes image decomposition and lighting inconsistency. By exploiting inherent discrepancies in imaging environments between genuine and fabricated images, this method extracts robust lighting cues and mitigates disturbances from environmental factors, revealing deeper-level alterations. A crucial element is the lighting information feature extractor, designed according to color constancy principles, to identify inconsistencies in lighting conditions. To address lighting variations, we employ a face material feature extractor using Pattern of Local Gravitational Force (PLGF), which selectively processes image patterns with defined convolutional masks to isolate and focus on reflectance coefficients, rich in textural details essential for forgery detection. Utilizing the Lambertian lighting model, we generate lighting direction vectors across frames to provide temporal context for detection. This framework processes RGB images, face reflectance maps, lighting features, and lighting direction vectors as multi-channel inputs, applying a cross-attention mechanism at the feature level to enhance detection accuracy and adaptability. Experimental results show that our proposed method performs exceptionally well and is widely applicable across multiple datasets, underscoring its importance in advancing deep forgery detection.

When we refer to “a green car,” it might sound as if that the color is an inherent characteristic of the object. However, color is a sensation produced by our visual system based on the light reflected from an object. Changes in scene illuminant alter the reflected light, so the color of the object should change as well. Yet, why does a green car parked in the morning still appear green when we return to it at dusk? This is due to color constancy, a phenomenon that maintains stable color perception despite changes in surrounding lighting environments. How does our visual system create a robust visual world from highly variable sensory signals? This article explains the mechanisms supporting human color constancy.

The visual fidelity of virtual reality (VR) and augmented reality (AR) environments is essential for user immersion and comfort. Dynamic lighting often leads to chromatic distortions and reduced clarity, causing discomfort and disrupting user experience. This paper introduces an AI-driven chromatic adjustment system based on a modified U-Net architecture, optimized for real-time applications in VR/AR. This system adapts to dynamic lighting conditions, addressing the shortcomings of traditional methods like histogram equalization and gamma correction, which struggle with rapid lighting changes and real-time user interactions. We compared our approach with state-of-the-art color constancy algorithms, including Barron’s Convolutional Color Constancy and STAR, demonstrating superior performance. Experimental results from 60 participants show significant improvements, with up to 41% better color accuracy and 39% enhanced clarity under dynamic lighting conditions. The study also included eye-tracking data, which confirmed increased user engagement with AI-enhanced images. Our system provides a practical solution for developers aiming to improve image quality, reduce visual discomfort, and enhance overall user satisfaction in immersive environments. Future work will focus on extending the model’s capability to handle more complex lighting scenarios.

Large efforts have been made to perform illuminant estimation, resulting in the development of various statistical- and learning-based methods. However, there have been challenges for some types of images, such as a single color, referred to as pure color images, which is the focus of the present research. In this study, the neural network approach is used. It was found the Kolmogorov-Arnold Networks (KAN) model, a novel approach that diverges from traditional Multi-Layer Perceptron (MLP) architectures gave the accurate predictions. Our method, ”Large Size Colour Constancy” (LSCC), characterized by its unique neural network structure, achieves high accuracy in illuminant estimation with significantly fewer parameters and enhanced interpretability. Additionally, three new pure color image datasets—”ZJU Color Fabric”, ”ZJU 0.8 Real Scene”, and ”ZJU 1.0 Real Scene” were produced—covering a wide range of conditions, including indoor and outdoor environments, as well as natural and artificial light sources. The results showed LSCC method to outperform existing methods across not only the pure colour datasets but also the traditional datasets, including classical normal images. It should offers practical deployment potential due to its efficiency and reduced computational requirements.

For given tristimulus values X, Y, Z of the object with reflectance ρ(λ) viewed under an illuminant S(λ) with tristimulus values X 0, Y 0, Z 0, an earlier algorithm constructs the smoothest metameric estimate ρ 0(λ) under S(λ) of ρ(λ), independent of the amplitude of S(λ). It satisfies a physical property of ρ(λ), i.e., 0≤ρ 0(λ)≤1, on the visual range. The second inequality secures the condition that for no λ the corresponding patch returns more radiation from S(λ) than is incident on it at λ, i.e., ρ 0(λ) is a fundamental metameric estimate; ρ 0(λ) and ρ(λ) differ by an estimation error causing perceptual variables assigned to ρ 0(λ) and ρ(λ) under S(λ) to differ under the universal reference illuminant E(λ)=1 for all λ, tristimulus values X E, Y E, Z E. This color constancy error is suppressed but not nullified by three narrowest nonnegative achromatic response functions A i(λ) defined in this paper, replacing the cone sensitivities and invariant under any nonsingular transformation T of the color matching functions, a demand from theoretical physics. They coincide with three functions numerically constructed by Yule apart from an error corrected here. S(λ) unknown to the visual system as a function of λ is replaced by its nonnegative smoothest metameric estimate S 0(λ) with tristimulus values made available in color rendering calculations, by specular reflection, or determined by any educated guess; ρ(λ) under S(λ) is replaced by its corresponding color R 0(λ) under S 0(λ) like ρ(λ) independent of the amplitude of S 0(λ). The visual system attributes to R 0(λ)E(λ) one achromatic variable, in the CIE case defined by y(λ)/Y E, replaced by the narrowest middle wave function A 2(λ) normalized such that the integral of A 2(λ)E(λ) over the visual range equals unity. It defines the achromatic variable ξ 2, A(λ), and ξ as described in the paper. The associated definition of present luminance explains the Helmholtz-Kohlrausch effect in the last figure of the paper and rejects CIE 1924 luminance that fails to do so. It can be understood without the mathematical details.

The visual system adapts dynamically to stabilize perception over widely varying illuminations. Such adaptation allows the colors of objects to appear constant despite changes in spectral illumination. Similarly, the wearing of colored filters also alters spectral content, but this alteration can be more extreme than typically encountered in nature, presenting a unique challenge to color constancy mechanisms. While it is known that chromatic adaptation is affected by surrounding spatial context, a recent study reported a gradual temporal adaptation effect to colored filters such that colors initially appear strongly shifted but over hours of wear are perceived as closer to an unfiltered appearance. Presently, it is not clear whether the luminance system adapts spatially and temporally like the chromatic system. To address this, spatial and temporal adaptation effects to a colored filter were measured using tasks that assess chromatic and luminance adaptation separately. Prior to and for 1 hour after putting on a pair of colored filters, participants made achromatic and heterochromatic flicker photometry (HFP) settings to measure chromatic and luminance adaptation, respectively. Results showed significant chromatic adaptation with achromatic settings moving closer to baseline settings over 1 hour of wearing the filters and greater adaptation with spatial context. Conversely, there was no significant luminance adaptation and HFP matches fell close to what was predicted photometrically. The results are discussed in the context of prior studies of chromatic and luminance adaptation.

Digital images (photos) have become an integral part of everyday life. This is supported by the increasingly sophisticated technology today where communication devices such as mobile phones can also use their role to replace analog cameras to take pictures or even record videos. Capturing happy moments or just wanting to record an event has become very easy to do, but sometimes the resulting photos are less than satisfactory due to the specifications of the mobile phone itself or other factors such as poor lighting (dark). Improvement efforts are also very necessary, but because the application to do this is only available on computers and not too many on mobile phones, this becomes difficult to do. One method in image improvement is the Histogram Equalization method. This method can be used to improve image quality related to lighting, namely by maintaining color constancy. The use of this histogram equalization method is considered easy because of its efficiency and relatively better performance on almost all types of images. The operation of HE (Histogram Equalization) is carried out by remapping the gray image level based on the probability distribution of the input gray level. It horizontally and dynamically stretches the various histograms of the image and produces an overall increase in contrast.

This paper shows how poor lighting can severely affect movies and influence the performance of video object detection systems and their practical applicability. This research problem was solved with the help of the proposed image enhancement model aimed at increasing the visibility and quality of low-illumination images. Hence, this study aims to help improve video object detection, particularly in regions of low illumination, utilizing the Enhanced Zero DCE model. This deep-learning framework does not require any reference images; therefore, it is appropriate for real-time applications. Enhanced Zero DCE eliminates DCE on high-order tonal curves, whereas the deep neural network boosts pixel values, resulting in better quality images. Using a variety of loss functions, including color constancy, exposure matching, smoothness, and spatial consistency, the model was deployed in the LoL dataset, which includes both high- and low-illumination pictures. From the experimental findings, a drastic enhancement in image quality improvement performance was evident. In terms of numbers, the effectiveness of the proposed model is outlined as follows: there is a gain of 23 percent luminance, 17 percent average illumination, 20 percent of the histogram mean, and illumination on objects when compared to traditional methods. These improvements were evident in an increase in detection accuracy by 15% and precision by 20%. Therefore, it can be said that the integration of the advanced Enhanced Zero DCE model significantly increases the efficiency of video detection of objects under dim-light conditions. These enhancements have practical applications in surveillance, automobiles, and other real-time video monitoring, especially in situations where accurate detection of objects is paramount.

GC[formula omitted]: Grouped convolutional color constancy

Flower colour polymorphisms are uncommon but widespread among angiosperms and can be maintained by a variety of balancing selection mechanisms. Anemone palmata is mostly yellow-flowered, but white-flowered plants coexist in some populations. We analysed the distribution of colour morphs of A. palmata across its range. We also characterised their colours and compared their vegetative and sexual reproductive traits, pollinator attention and fitness. The range of A. palmata is limited to the Western Mediterranean, while white-flowered plants are restricted to Portugal and SW Spain, where they occur at low proportions. Yellow flowers have a characteristic UV pattern, with a UV-absorbing centre and UV-reflecting periphery, which is absent in the white morph. Colour features of both morphs were highly delineated, making it easy for pollinators to distinguish them. Both morphs were protogynous, with the same duration of sexual stages, and the main floral traits related to pollinator attraction, apart from flower colour, were similar. Hymenoptera and Diptera were the main pollinators, showing preference for the yellow morph, clear partitioning of pollinator groups between the two colour morphs and a marked constancy to flower colour during foraging. Both morphs combined clonal propagation with sexual reproduction, but sexual reproductive potential was lower in white-flowered plants. Finally, female fitness was higher in the yellow morph. Pollinator partitioning and colour constancy could maintain this polymorphism, despite the lower visitation rate and fitness of white-flowered plants, which could facilitate their clonal propagation.