Fig. 1. 3D integral imaging. (a) Pickup stage^[12]. (b) Reconstruction stage^[12]. (c) 3D underwater computational reconstruction^[18]. Reprinted with permissions from Refs. [12] and [18].

Fig. 2. (a) Clear water 3D scene used in the experiments. (b) The scene in turbid water. (c) Diagram of the 3D underwater imaging system in turbid water. (d)–(g) 3D integral imaging reconstruction. (d) 3D reconstructed bug focused at zr=490mm, (e) 3D reconstructed treasure focused at zr=640mm, (f) 3D reconstructed small fish focused at zr=730mm, and (g) 3D reconstructed large fish focused at zr=770mm. Reprinted with permission from Ref. [18].

Fig. 3. Flow chart of peplography for imaging in scattering media. Reprinted with permission from Ref. [19].

Fig. 4. (a) Sample clean image of a light source (α=0.0025mm−1). (b)–(c) Degraded images in high turbid water (α=0.040mm−1). (d) Recovered clean image from (b). (e) Recovered clean image from (c). α is Beer’s coefficient. Reprinted with permission from Ref. [21].

Fig. 5. (a) Sample clean image of a light source (α=0.0025mm−1). (b) 2D degraded central perspective image in turbidity, ambient light, and partial occlusion (α=0.008mm−1). (c) Recovered clean 2D image from (b). (d) Recovered clean 3D reconstructed image from (b). α is Beer’s coefficient. Reprinted with permission from Ref. [21].

Fig. 6. Flow chart of the underwater optical signal detection pipeline using 4D nonlinear correlation. Reprinted with permission from Ref. [12].

Fig. 7. Flow chart of CNN-BiLSTM-based signal detection using sliding window-based classification. Reprinted with permission from Ref. [13].

Fig. 8. Flow chart of end-to-end integrated 1DInImCNN-based signal detection model. Reprinted with permissions from Refs. [14,15].

Fig. 9. (a)–(c) Sample training data without occlusion. (a) Central perspective video of the encoded optical signal in clear water. (b)–(c) Sample training data in turbid water from a lensless-based 1D camera array with diffusers at α=0.0037mm−1 and α=0.0095mm−1, respectively. (d)–(f) Sample testing data with occlusion. (d) Central perspective video of the encoded optical signal in clear water with partial occlusion. (e)–(f) Sample testing data in turbid water with partial occlusion from a lensless-based 1D camera array with diffusers at α=0.0170mm−1 and α=0.0198mm−1, respectively. Reprinted with permission from Ref. [15].

Fig. 10. Sample underwater testing scenes with occlusion applying object detection and signal detection. Reprinted with permission from Ref. [16].