带有光照引导和分数阶梯度的变分Retinex模型用于低光照图像增强
Variational Retinex model with illumination guidance and fractional derivative for low-light enhancement
在获得图像的过程中,不足的环境光照会导致低光照图像的产生。这种退化现象表现为图像具有较低的亮度对比度和部分噪声,产生颜色失真和细节模糊。这一退化不仅会直接影响观看者的直观体验,更深层次地,会影响以图像或视频作为输入的更高层次的计算机视觉任务。对低光照图像进行增强,要避免产生过度曝光。与此同时,也需要提升图像的对比度,消除噪声,保留图像的自然性。这是一项具有重要意义,也充分具有挑战性的任务。本文提出了一种变分Retinex模型。该模型通过结构/纹理感知映射(structure/texture aware map, SAM/TAM), 将输入图像的经典/分数阶导数引入到光照和反射部分的正则项中,以提高Retinex分解和之后光照增强的效果。通过对输入图像的经典（分数阶）导数的平方倒数,生成了SAM(TAM)。为了提高分解效果,并将光照引导项合并到目标函数中。光照引导项通过惩罚光照与通过暗通道先验和引导图像滤波预估的光照之间的差异,促使光照部分包含更多的结构信息。最后,采用一种交替算法来解决模型中涉及的最小化问题。评估结果显示,所提出的方法在低光照增强方面表现更好。
Insufficient ambient light during image capture can lead to low-light image generation, characterized by diminished brightness contrast, partial noise, color distortion, and blurred details. This degradation not only impacts viewer experience but also affects higher-level computer vision tasks relying on images or videos as input.Enhancing low-light images requires avoiding overexposure while improving contrast, eliminating noise, and preserving naturalness. Our proposed Variational Retinex model leverages Structure/Texture Aware Maps (SAM/TAM) to incorporate classical/fractional derivatives, enhancing Retinex decomposition and subsequent illumination improvement. Evaluation results indicate superior performance in low-light enhancement.
邓尚玮、何传江
光电子技术电子技术应用计算技术、计算机技术
计算数学低光照图像增强Retinex分解分数阶梯度引导滤波稳健统计
computational mathematicsLow-light enhancementRetinex decompositionFractional differentiationGuided filteringRobust statistics
邓尚玮,何传江.带有光照引导和分数阶梯度的变分Retinex模型用于低光照图像增强[EB/OL].(2024-03-20)[2025-08-16].http://www.paper.edu.cn/releasepaper/content/202403-275.点此复制
评论