Fig. 1. Flow chart of the proposed algorithm

Fig. 2. Image decomposition

Fig. 3. Block effects and halo effects. (a) Original image; (b) dark channel priors estimate transmittance; (c) restored image

Fig. 4. Comparison of point transmission map. (a) Transmission estimated by pixel-based dark color; (b) result of Ref. [18]; (c) transmission estimated by proposed method; (d) dehazing result by proposed method

Fig. 5. Flow chart of transmission fusion

Fig. 6. Structure of decision image

Fig. 7. Hazy images and their corresponding decision images. (a)(d) Original images; (b)(e) histograms of decision images; (c)(f) decision images

Fig. 8. Hazy image and its corresponding decision value of 0.1% pixel for the dark channel. (a)(c) Original images; (b)(d) histograms of dark channel decision image

Fig. 9. Transmission optimization results with different filtering radius

Fig. 10. Filtering with the original image as the guided image

Fig. 11. Processes of multi-guided filtering

Fig. 12. Comparison results. (a) Original images; (b) method in Ref. [18]; (c) proposed method

Fig. 13. Comparison results. (a) Original images; (b) method in Ref. [21]; (c) proposed method

Fig. 14. Comparison results. (a) Original images; (b) method in Ref. [22]; (c) proposed method

Fig. 15. Comparison results of proposed method and traditional methods. (a) Original images; (b) method in Ref. [11]; (c) method in Ref. [6]; (d) method in Ref. [7]; (e) proposed method

Fig. 16. Comparisonof results of proposed method and deep learning methods. (a) Original images; (b) method in Ref. [13]; (c) method in Ref. [12]; (d) proposed method