Underwater imaging is significantly affected by scattering and absorption caused by suspended particles in the water, leading to a considerable decline in image quality. Existing polarization-based dehazing algorithms primarily focus on estimating backscattered light parameters during image restoration. However, underwater imaging scenarios, such as nighttime or deep-sea environments, often require active light sources for illumination. The non-uniform distribution of these active light sources results in uneven image brightness, complicating the estimation of backscattered light. This challenge makes it difficult to fully suppress backscattered light, which adversely affects the restoration process. In addition, non-uniform illumination can cause overexposure or underexposure in captured images, leading to the loss of critical object information. This hampers subsequent tasks such as underwater detection, object recognition, and object tracking. To address these challenges, we propose a novel image restoration method that leverages polarimetric imaging to handle non-uniform illumination. The method employs low-rank sparse matrix decomposition to separate backscattered light from object information light. Subsequently, backscattered light is corrected and combined with an underwater imaging model to achieve clarity and uniformity in the restored images. The proposed approach holds significant potential for applications in underwater polarization imaging, object recognition, and tracking.

The proposed method focuses on the difference in matrix dimensions between object information light and backscattered light during underwater imaging. Initially, four images with varying polarization angles, captured using a split-focus plane camera, are preprocessed. Polarized images at any given angle are represented using Stokes vectors. Based on the regional detail richness of the images, two polarization sub-images are selected for processing. Since backscattered light exhibits low-rank characteristics while object information light is sparse, the method uses low-rank sparse matrix decomposition to achieve preliminary separation. The backscattered light is then corrected for uniformity through adaptive brightness adjustment, Gamma correction, and local brightness equalization. Using an underwater scattering model, the intensity of backscattered light at infinity and the transmittance map are estimated to restore the two polarized sub-images. Finally, these restored images are fused to produce a clear underwater image.

Polarization imaging experiments are conducted in turbid water at various concentrations under non-uniform illumination conditions (Fig. 3). The results demonstrate that the proposed method effectively corrects non-uniform illumination and significantly enhances the clarity and contrast of underwater images. Using light emitting diode (LED) lighting as an active illumination source, experiments are performed in water bodies with low, medium, and high turbidity levels. Comparisons with other descattering methods reveal superior results, with uniformly distributed brightness and no overexposure. The method also significantly enhances contrast and clarity (Fig. 6). Partial zoomed-in views reveal richer image details, particularly as water turbidity increases, highlighting the method’s advantages (Fig. 7). In addition, three-dimensional grayscale analyses before and after restoration further verify the method’s effectiveness in correcting non-uniform illumination. Compared with the original image, pixel grayscale values in the restored images exhibit a more even distribution (Fig. 8). A quantitative analysis using five image quality metrics, eight-neighbor contrast, enhancement measure evaluation (EME), average gradient (AG), edge intensity (EI), and underwater image quality measures (UIQM), further confirms the proposed method’s effectiveness and superiority (Table 1).

To address the degradation of underwater image quality caused by non-uniform illumination and scattering effects, we propose a restoration method based on low-rank sparse matrix decomposition and illumination correction. Two polarized sub-images are selected based on image contrast, and low-rank sparse matrix decomposition is applied to separate backscattered light from object information light. After homogenizing the backscattered light, parameters such as the transmittance map and backscattered light at infinity are estimated. These are combined with an underwater scattering model to restore the two polarized sub-images. The restored images are fused to produce a clear and enhanced underwater image. Multiple experiments are conducted in turbid water with different concentrations under non-uniform illumination methods. The proposed method’s results are qualitatively and quantitatively compared with other descattering techniques. The findings demonstrate that the method effectively corrects non-uniform illumination and significantly improves image contrast and detail richness, resulting in enhanced image quality.