Author Affiliations
1College of Information, Shanghai Ocean University, Shanghai 201306, China2Shanghai University of Electric Power, Shanghai 201306, China3Institute of Deep-Sea Science and Engineering, Chinese Academy of Sciences, Sanya 572000, Hainan , Chinashow less
Fig. 1. Flow chart of constructing the underwater video quality assessment dataset
Fig. 2. Examples of five classes of contents about underwater videos. (a) Submarine rubbish; (b) submarine topography; (c) hydrothermal vents; (d) marine operation; (e) marine life
Fig. 3. Original and Fusion enhanced underwater images. (a) Original image; (b) image enhanced with Fusion model
Fig. 4. Original and Ucolor enhanced underwater images.(a) Original image; (b) image enhanced with Ucolor model
Fig. 5. ROI area (inside the box) and non-ROI area (outside the box) of a single frame
Fig. 6. Flow chart of subjective assessment of video quality
Fig. 7. Subjective video quality assessment system. (a) Viewing interface; (b) assessment interface
Fig. 8. Performance of underwater enhancement models in low light environment. (a) Original frame, MOS is 28.1; (b) enhanced by Fusion, MOS is 17.2; (c) enhanced by Ucolor, MOS is 20.9
Fig. 9. Performance of underwater enhancement models in color cast environment. (a) Original frame, MOS is 43.6; (b) enhanced by Fusion, MOS is 58.4; (c) enhanced by Ucolor, MOS is 50.4
Type of degradation | Parameter | ROI | Non-ROI |
---|
Gaussian blurring | Kernel | | | Sigma | 3 | 1 | Gaussian noise | Mean | 0 | 0 | STD | 11 | 8 |
|
Table 1. Details of video quality degradation parameters
Method | Enhancement model | Degradation model |
---|
Ucolor | Fusion | ROI_GB | ROI_GN |
---|
Percentage of quality scores higher or lower than the original video /% | 35 | 33 | -72 | -67 |
|
Table 2. Subjective performance of video quality enhancement and degradation methods
Classification of objective quality assessment model | Model | Correlation index |
---|
PLCC | SROCC |
---|
Underwater scene | Image | UCIQE[22] | 0.3259 | 0.2293 | Image | UIQM[23] | 0.3054 | 0.3272 | Image | Guo’s[24] | 0.4310 | 0.3219 | Video | Moreno-Roldán’s[9] | 0.3455 | 0.2590 | Video | Song’s[6] | 0.5103 | 0.4936 | Terrestrial scene | Image | Brisque[25] | 0.3953 | 0.4239 | Video | VBliinds[27] | 0.6735 | 0.6336 | Video | VIIDEO[26] | 0.6423 | 0.6060 |
|
Table 3. Performance comparison of different image/video quality assessment models on the dataset constructed in this paper
Objective assessment model | PLCC/SROCC |
---|
Ours | Song’s dataset[6] | Moreno-Roldán’s dataset[8] |
---|
Song’s[6] | 0.51/0.49 | 0.84/0.83 | — | Moreno-Roldán’s[9] | 0.35/0.26 | — | 0.80/0.76 | VIIDEO[26] | 0.64/0.61 | 0.01/0.01 | 0.12/0.11 |
|
Table 4. Results of video quality assessment models on different underwater video datasets
Feature class | Feature name | Correlation |
---|
Spatial domain | Underwater image colorfulness(UIQM) | 0.299** | Underwater image contrast(UIQM) | 0.284** | Natural image statistics | Brisque_1(Brisque) | -0.369** | Brisque_2,3,4,6,8,10,12,14,16,18(Brisque) | -0.230**~-0.312** | NIQE_4(VBliinds) | 0.260** | NIQE_5(VBliinds) | -0.216** | NIQE_8(VBliinds) | 0.247** | NIQE_16(VBliinds) | 0.292** | NIQE_22(VBliinds) | 0.216** | Frequency domain | DC_variation(VBliinds) | 0.430** | Motion | Global motion(VBliinds) | -0.249** |
|
Table 5. Correlation analysis of video quality characteristics