基于深度学习的低照度图像增强方法

Low-light image enhancement method based on deep learning

在低照度环境下采集的图像往往亮度不足,导致在后续视觉任务中难以有效利用。针对这一问题,过去的低照度图像增强方法大多在极度低光场景中表现失败,甚至放大了图像中的底层噪声。为了解决这一难题,本文提出了一种新的基于深度学习的端到端神经网络,该网络主要通过空间和通道双重注意力机制来抑制色差和噪声,其中空间注意力模块利用图像的非局部相关性进行去噪,通道注意力模块用来引导网络细化冗余的色彩特征。实验结果表明,与其他主流算法相比,本文方法在主观视觉和客观评价指标上均得到了进一步提高。

Images acquired in low-light environments are often not bright enough,making them difficult to use effectively in subsequent visual tasks.In response to this problem,most of the past low-light image enhancement methods have failed in extreme low-light scenes and even magnified the underlying noise in the image.In order to solve this problem,this paper proposes a new end-to-end neural network based on deep learning,which is primarily based on spatial and channel dual attention mechanism to suppress chromatic aberration and noise.The spatial attention module uses the non-local correlation of the image for denoising,and the channel attention module is used to guide the network to refine the redundant color features.The experimental results show that the method in this paper is further improved in both subjective visual and objective evaluation metrics compared to other mainstream algorithms.