grayscale image preprocessing

Semantic segmentation performance degrades significantly under out-of-distribution adverse conditions—such as nighttime, fog, and snow—due to degraded visual cues. To address this, we propose a biologically inspired, lightweight input preprocessing method that mimics the human retina’s contrast enhancement (via Difference-of-Gaussians filtering applied to RGB, grayscale, and opponent-color channels) and color antagonism mechanisms. Our approach is plug-and-play: it requires no architectural modifications or model retraining. It is model-agnostic and computationally efficient. Evaluated on Cityscapes, ACDC, and Dark Zurich, the method preserves accuracy on in-distribution data (e.g., clear daytime scenes) while substantially improving robustness and generalization stability under challenging conditions—including low illumination, fog, and snow. This establishes an effective and reliable preprocessing paradigm for semantic segmentation in open-world environments.

This paper addresses three critical challenges in high-resolution image processing: grayscale quantization distortion, low feature extraction accuracy, and non-invertible transformations. To this end, we propose a modular spatial image processing framework. Methodologically, it employs a multi-level collaborative algorithm architecture integrating structure-preserving 8-level grayscale discretization, RGB/YCrCb histogram equalization, HSV brightness adaptive adjustment, 3×3 convolution-based sharpening and unsharp masking, gamma correction, and joint extraction of Canny edges, Hough lines, Harris corners, and morphological geometric features. Additionally, a reversible bidirectional transformation pipeline is designed to ensure consistency between forward processing and inverse reconstruction. Experimental results demonstrate a bidirectional transformation fidelity of 76.10% (forward) and 74.80% (inverse), cue stick angle estimation error below 0.5° (measured at 51.50°), cue isolation similarity to the original image of 81.87%, and strong robustness across multiple benchmark datasets.

This work addresses the inherent ambiguity in grayscale image colorization, which often leads to semantically inaccurate color predictions. To mitigate this issue, the authors propose incorporating CLIP-based textual conditioning as a guidance signal and present the first systematic evaluation—under controlled conditions—of how text prompts influence colorization performance across two distinct architectures: a U-Net and Stable Diffusion 1.5. Experimental results demonstrate that text guidance substantially enhances colorization quality: for the U-Net, PSNR improves by 5.6%, SSIM by 1.2%, colorfulness by 36.6%, and LPIPS decreases by 7.6%. Consistent improvements are also observed with Stable Diffusion, confirming the effectiveness and generalizability of text-guided colorization across different model architectures.

This work addresses the growing security risks posed by the proliferation of generative models on resource-constrained smart devices by proposing a lightweight and efficient method for detecting synthetic images. The approach employs a minimal preprocessing pipeline based on pixel-wise gradient computation, leveraging high-pass filtering to emphasize local grayscale intensity variations—identified as a key discriminative cue—while simultaneously suppressing color-related artifacts. Without relying on complex architectures or substantial computational resources, the proposed paradigm achieves detection accuracy comparable to state-of-the-art methods across multiple benchmark datasets, yet with significantly reduced computational overhead, thereby enabling real-time deployment on devices such as smartphones.

Existing image recoloring and editing methods struggle to simultaneously achieve precise local color control, faithful texture preservation, and consistent color reproduction across luminance-varying regions. To address this, we propose a diffusion-based, quantized palette-driven editing framework. Our method uniquely employs quantized images as direct inputs to the diffusion model, enhancing interpretability and controllability. We design a weighted bipartite graph matching algorithm to enable semantically coherent, extreme palette transfer. Furthermore, we introduce multi-scale texture conditioning—optimized via thresholded gradient guidance—and JPEG-noise-robust training, overcoming the brightness-invariance limitation inherent in prior approaches. Extensive experiments demonstrate state-of-the-art performance: high-fidelity reconstruction, strict adherence to target palettes, and superior texture consistency. The framework supports both localized recoloring with fine-grained control and end-to-end palette migration, establishing a new benchmark for controllable, photorealistic image editing.

This work addresses the growing challenge posed by increasingly photorealistic generative AI images to image forensics. It proposes, for the first time, a systematic and interpretable color transformation framework that exploits statistical discrepancies in chrominance between real and synthetic images. The approach integrates six handcrafted and one task-driven learned color transformations to extract color-sensitive features at either pixel or patch levels, which are then fed into a lightweight classifier for high-accuracy detection. The method achieves an average generalization accuracy of 93.27% under various post-processing attacks, enables intuitive authenticity assessment through visualizable noise patterns, and significantly enhances source attribution across diverse generative models.

This work addresses the insufficient accuracy of the Oklab/Oklch color spaces in predicting perceptual color differences, which limits their utility in high-consistency design and interpolation tasks. The authors propose Oklch+, an enhanced color space incorporating a power transformation along the lightness (L) axis and Naka–Rushton–based saturation compression along the chroma (C) axis, yielding a model with only three tunable parameters. Perceptual color differences are approximated using Euclidean distance in this modified space. Optimized on the COMBVD dataset, Oklch+ achieves a STRESS value of 29.09—comparable to CIEDE2000 (29.13)—and further improves to 26.14 on the BFD-P D65 holdout set, substantially outperforming original Oklab (51.45). Thus, Oklch+ approaches the predictive performance of CIEDE2000 while retaining simplicity and interpolability.

This work addresses the severe color distortion commonly observed in low-light raw image denoising, which is primarily caused by black level errors—a factor previously overlooked in the literature. The study is the first to identify black level error as the dominant source of color shifts in such scenarios and introduces a blind denoising method that requires neither camera calibration nor device-specific parameters, rendering it universally applicable. The proposed approach employs a global bias estimation network to correct black level errors and establishes an unbiased ground-truth extraction framework to rectify color biases in the SIDD dataset. Without paired training data or camera metadata, the deep network achieves high-fidelity color-preserving denoising, significantly outperforming existing blind methods on the ELD, SID, and LRID benchmarks, with certain metrics even matching or surpassing those of strongly supervised approaches.

This work addresses the challenge of high model complexity and excessive parameter count in low-light image enhancement by proposing a lightweight two-stage enhancement framework. The method integrates a frozen distribution normalization preprocessing module—which provides luminance correction priors—with a compact U-Net architecture built entirely from depthwise separable convolutions to perform residual color correction. This design significantly reduces model parameters while preserving excellent perceptual quality. The proposed approach secured fourth place in the CVPR 2026 NTIRE Challenge on Efficient Low-Light Image Enhancement and demonstrates strong effectiveness and generalizability through comprehensive benchmarking and ablation studies.

This study addresses the critical issue of racial bias in chest X-ray (CXR) diagnostic models, which often exploit implicit racial cues embedded in images, thereby compromising healthcare equity. The authors systematically evaluate the impact of various image preprocessing techniques—including lung masking, lung cropping based on bounding boxes, and Contrast-Limited Adaptive Histogram Equalization (CLAHE)—on mitigating race-encoded information in CXRs. Their findings demonstrate that bounding box–based lung cropping substantially reduces the model’s reliance on racial shortcuts while preserving high diagnostic accuracy for pulmonary diseases. This approach effectively circumvents the conventional trade-off between fairness and performance, offering a promising pathway toward developing more reliable and equitable AI systems in medical imaging.