基于曲线拟合与去噪的弱光图像增强算法

Low-light Image Enhancement Algorithm Based on Curve Estimation and Denoising

在弱光、背光、非均匀光照等干扰下,弱光图像增强算法很难获得优质的图像增强效果。论文针对这些干扰引起的噪声与细节信息丢失等问题,提出了一种基于曲线拟合与去噪的弱光图像增强算法。该算法主要由曲线拟合、分解、去噪和优化四个子网络构成,通过感知损失和细节损失等损失函数来对网络进行约束,最终得到在对比度、细节、颜色、噪声等方面效果更好的增强图像。此外,在训练阶段,该方法加入了一个残差模块使得深层神经网络在训练阶段更容易优化,并且不会导致梯度消失或爆炸。在数据集LOL上的实验结果表明,该算法在图像质量指标峰值信噪比和结构相似性上获得了良好的性能。特别地,与基准算法Zero-DCE相比,该算法在PSNR和SSIM指标上更是提升了14.7%和32.8%。

Under the interference of low light,backlight and non-uniform light,it is difficult to obtain high-quality image en-hancement.To relieve the issue caused by the problem mentioned above,this paper proposes a low-light image enhancement algo-rithm based on curve fitting and denoising.The network is mainly composed of four sub networks:curve estimation,decomposition,denoising and optimization.Furthermore,the network is supervised by the perceptual loss and detail loss,which is the main loss function.Afterwards,a result is obtained by the decoder module,which has a better performance in contrast,detail,color,noise and so on.Moreover,a residual learning module is employed to make the deep neural network more easier to optimize in the training stage,and will alleviate the problems that caused by the gradient disappearing or explosion.The experimental results on LOL datas-et show that the algorithm achieves better results than other compared solutions in the metrics of peak signal-to-noise ratio and structural similarity.Compared with our baseline algorithm Zero-DCE,the proposed method has better performance on both metrics of PSNR and SSIM,which have huge gains of 14.7%and 32.8%,respectively.