基于深度学习的矿井下低照度图像增强算法研究

Research on image enhancement algorithm of low illumination in mine based on deep learning

针对传统图像增强算法在处理井下图片时存在的噪声大、图像颜色失真等问题,提出一种基于深度学习的KinD-Net算法应用于矿井下环境对图像进行增强。该算法在图像分解网络中加入重构损失,反射图恢复网络中加入去噪模块,抑制图像增强过程中产生的噪声。在反射恢复网络中引入照度图信息减少图像颜色失真。将本文算法与Retinex算法、MSRCR算法、HSV MSRCR算法与Retinex-Net算法进行对比。实验结果表明,本文算法峰值信噪比、机构相似性方面表现较优,可以有效提高图像的亮度,优化图像颜色失真问题。

This paper proposes a KinD-Net algorithm based on deep learning to enhance images in underground environment,aiming at the problems of noise and color distortion of traditional image enhancement algorithm in processing underground images.In this algorithm,reconstruction loss is added into the image decomposition network and denoising module is added into the reflection image restoration network to suppress the noise generated in the image enhancement process.Introduce illuminance map information to reduce color distortion in reflection recovery network.The proposed algorithm is compared with Retinex algorithm,MSRCR algorithm,HSV MSRCR algorithm and Retinex-Net algorithm.Experimental results show that the proposed algorithm performs better in terms of PSNR and SSIM,and can effectively improve image brightness and optimize image color distortion.