Fig. 1. MCIEN model

Fig. 2. Example images in MIT-Adobe 5K dataset. (a) Low-light images; (b) normal images

Fig. 3. Enhanced results of the first group of images. (a) Underexposed image; (b) LIME; (c) LLNet; (d) RetinexNet; (e) MBLLEN; (f) EGAN; (g) DSLR; (h) MIRNet; (i) KinD++; (j) ZDCE; (k) proposed method; (l) expert-retouched image

Fig. 4. Enhanced results of the second group of images. (a) Underexposed image; (b) LIME; (c) LLNet; (d) RetinexNet; (e) MBLLEN; (f) EGAN; (g) DSLR; (h) MIRNet; (i) KinD++; (j) ZDCE; (k) proposed method; (l) expert-retouched image

Fig. 5. Enhanced results of the third group of images. (a) Underexposed image; (b) LIME; (c) LLNet; (d) RetinexNet; (e) MBLLEN; (f) EGAN; (g) DSLR; (h) MIRNet; (i) KinD++; (j) ZDCE; (k) proposed method; (l) expert-retouched image

Fig. 6. Local amplification of the second group of enhanced images. (a) Local amplification of Fig.4(e), MBLLEN; (b) local amplification of Fig.4(h), MIRNet; (c) local amplification of Fig.4(k), proposed method

Fig. 7. Histogram comparison results of the first group of enhanced images. (a) Histogram of Fig.3(e), MBLLEN; (b) histogram of Fig.3(h), MIRNet; (c) histogram of Fig.3(k), proposed method

Fig. 8. Comparison of network model. (a) w/o SK; (b) one SK; (c) w/o CA

Fig. 9. Examples of enhanced image acquired under extreme conditions. (a) Sunrise; (b) Sun_en; (c) Mountain; (d) Moun_en