Correcting Low-Illumination Images Using Multi-Scale Fusion in a Pyramidal Framework

Existing multi-exposure fusion (MEF) algorithms for gray images under low-illumination cannot preserve details in dark and highlighted regions very well, and the fusion image noise is large. To address these problems, an MEF method is proposed. First, the latent low-rank representation (LatLRR) is used on low-dynamic images to generate low-rank parts and saliency parts to reduce noise after fusion. Then, two components are fused separately in Laplace multi-scale space. Two different weight maps are constructed according to features of gray images under low illumination. At the same time, an energy equation is designed to obtain the optimal ratio of different weight factors. An improved guided filtering based on an adaptive regularization factor is proposed to refine the weight maps to maintain spatial consistency and avoid artifacts. Finally, a high dynamic image is obtained by the inverse transform of low-rank part and saliency part. The experimental results show that the proposed method has advantages both in subjective and objective evaluation over state-of-the-art multi-exposure fusion methods for gray images under low-illumination imaging.

Download Full-text

Low illumination image enhancement based on multi-scale CycleGAN with deep residual shrinkage

Journal of Intelligent & Fuzzy Systems ◽

10.3233/jifs-211664 ◽

2021 ◽

pp. 1-13

Author(s):

Daxin Zhou ◽

Yurong Qian ◽

Yuanyuan Ma ◽

Yingying Fan ◽

Jianeng Yang ◽

...

Keyword(s):

Image Enhancement ◽

Image Features ◽

Generative Adversarial Networks ◽

Style Transfer ◽

Multi Scale ◽

Adversarial Networks ◽

Network Methods ◽

Convolution Kernels ◽

Low Illumination ◽

Noise Amplification

Low-illumination image restoration has been widely used in many fields. Aiming at the problem of low resolution and noise amplification in low light environment, this paper applies style transfer of CycleGAN(Cycle-Consistent Generative Adversarial Networks) to low illumination image enhancement. In the design network structure, different convolution kernels are used to extract the features from three paths, and the deep residual shrinkage network is designed to suppress the noise after convolution. The color deviation of the image can be resolved by the identity loss of CycleGAN. In the discriminator, different convolution kernels are used to extract image features from two paths. Compared with the training and testing results of Deep-Retinex network, GLAD network, KinD and other network methods on LOL-dataset and Brightening dataset, CycleGAN based on multi-scale depth residuals contraction proposed in this experiment on LOL-dataset results image quality evaluation indicators PSNR = 24.62, NIQE = 4.9856, SSIM = 0.8628, PSNR = 27.85, NIQE = 4.7652, SSIM = 0.8753. From the visual effect and objective index, it is proved that CycleGAN based on multi-scale depth residual shrinkage has excellent performance in low illumination enhancement, detail recovery and denoising.

Download Full-text