提示
多摄像机协同暗光图像增强算法
首发时间：2025-03-10
摘要：暗光图像因光照不足普遍存在亮度衰减、对比度下降等问题，导致视觉质量劣化与后续分析性能受限。而当前大部分暗光增强算法仅基于单一摄像机捕获的图像数据，在应对复杂光照退化时面临参照信息维度单一、细节恢复能力不足等问题。视联网中，多摄像机系统通过协同工作，为解决这一问题提供了有效的方案。通过整合捕获的多视角图像数据，可以获得更为全面的视觉信息，为暗光图像增强提供更丰富的语义、纹理、颜色等内容，提升增强后图质量。因此，本文提出了一种基于深度retinex理论的多摄像机协同暗光图像增强算法，并构建了一个包含500组成对暗光/正常光图像的多摄像机协同暗光图像增强数据集。算法分为图像分解、反射图光照图调整和增强图像重建三个阶段，引入特征点匹配损失函数和多视角融合模块以充分利用多视角信息。在本文拍摄的数据集和middlebury数据集上的实验结果验证了该算法的有效性。此外，本文还通过消融实验证明了特征点匹配损失函数和多视角融合模块对于模型的重要性，由此证明了算法设计的合理性。
关键词：
for information in english, please click here
multi-camera collaborative low-light image enhancement algorithm
abstract：low-light images, due to insufficient illumination, generally suffer from brightness attenuation and contrast reduction, leading to degraded visual quality and limited performance in subsequent analysis. current low-light enhancement algorithms mostly rely on image data captured by a single camera, and they face issues such as singular reference information dimension and insufficient detail recovery capability when dealing with complex illumination degradation.the multi-camera system in viot, through collaborative operation, offers an effective solution to this problem. the multi-camera system can capture the scene from different perspectives. by integrating the multi-view image data, it can obtain more comprehensive visual information, providing richer semantics, textures, and colors for low-light image enhancement and improving the quality of the enhanced images. therefore, this paper proposes a multi-camera collaborative low-light image enhancement algorithm based on the deep retinex theory and constructs a multi-camera collaborative low-light image enhancement dataset containing 500 pairs of low-light and normal-light images. the algorithm consists of three stages: image decomposition, reflectance and illumination adjustment, and enhanced image reconstruction. it introduces a feature matching loss function and a multi-view fusion module to fully utilize multi-view information. experimental results on the dataset captured in this paper and the middlebury dataset validate the effectiveness of the proposed algorithm. moreover, ablation studies demonstrate the importance of the feature matching loss function and the multi-view fusion module to the model, thereby proving the rationality of the algorithm design.
keywords：
基金：
论文图表：
引用
导出参考文献
no.****
同行评议
勘误表
多摄像机协同暗光图像增强算法
评论
全部评论