Fig. 1. Example of autoencoder (AE)

Fig. 2. Feature extraction network

Fig. 3. Dense block structure

Fig. 4. Structure of texture refinement. (a) Texture refinement network; (b) texture refinement unit

Fig. 5. Degraded underwater image, clear underwater image corresponding to degraded underwater image, HSV color space image corresponding to degraded underwater image and its component images. (a) Degraded images, (b) corresponding clear underwater images, (c) corresponding HSV color space images, (d) corresponding H component images, (e) corresponding S component images, and (f) corresponding V component images of coral and whale skeleton

Fig. 6. Flow chart of proposed algorithm

Fig. 7. Experimental results of different algorithms. (a) Original images; (b) CLAHE; (c) UDCP; (d) FE; (e) CycleGAN; (f) WSCT; (g) prposed algorithm

Fig. 8. Diagram of different models. (a) model2; (b) model3; (c) model4; (d) model5

Fig. 9. Different model's results. (a) model2's result; (b) model3's result; (c) model4's result; (d) model5's result; (e) proposed model's result

Fig. 10. Objective evaluation indicators for different number of dense blocks

Table 1. UICQE comparison of results of proposed method and several other algorithms

Table 2. UIQM comparison of results of proposed method and several other algorithms

Table 3. Subjective evaluation index comparison of results of proposed model and several other models