Scattering and absorption of light by suspended particles in scattering media can lead to significant degradation of image quality. In the imaging environment of turbid water, the interference of backscattered light leads to image degradation, and the light is partially polarized. In order to obtain clear images, it is necessary to suppress backscattered light, which is also the core of underwater polarization imaging technology. Most of the existing underwater polarization imaging methods distinguish scattered light from signal light in the spatial domain based on their polarization information. However, overlapping in the spatial domain makes it difficult to fully suppress the backscattered light only by using the polarization characteristic difference between the target signal light and the backscattered light. In fact, the two kinds of lights are distributed differently in the spectrum of a linearly polarized image. The backscattered light, which is the main cause of image degradation, is relatively concentrated in the low-frequency components of the spectrum, while the information of the target signal light is mainly distributed in the high-frequency components. In this paper, an underwater polarization imaging method based on polarization image spectrum processing is proposed, which effectively suppresses backscattered light by the frequency domain filtering and polarization correlation processing of orthogonally polarized images and improves the contrast and clarity of underwater images. The method proposed in this paper and the results are of great significance to the research on polarization image enhancement and underwater clear imaging.

The proposed method mainly utilizes the spectrum distribution differences between the target signal light and backscattered light. With the help of frequency domain filtering and polarimetric recovery, this method separates and suppresses scattered light in steps and finally achieves high-quality image recovery. First, the cross-linear image in orthogonal polarization images is processed by a high-pass filter, which contains more target information, so as to automatically search for the optimal cut-off frequency of the filter. By extracting high-frequency components, the target signal light and backscattered light are preliminarily separated. With the high-frequency components of the cross-linear image, polarization-related processing is performed on the co-linear image. In fact, there is a polarization relation between the two orthogonal polarization images, which can be represented by the degree of polarization (DOP). In order to maintain this intrinsic polarization relation, it is necessary to keep DOP constant and process the co-linear image based on the high-frequency component of the cross-linear image filtered by the optimal filter. Then the processed orthogonal polarization images are employed to estimate parameters including the DOP of backscattered light, the value of backscattered light from infinity in turbid media, and the transmission based on the improved traditional polarimetric recovery method, and then the recovered image of the target objects in underwater can be finally obtained.

The results of polarization imaging experiments of different objects in water with different turbidity show that the proposed method can efficiently restore the images of target objects and improve the underwater imaging quality. In experiments of the panda-pattern target, search results of the optimal cut-off frequency of each filter are visualized in Fig. 4. By analyzing the intensity histogram at the dash of the result images (Fig. 5), it is found that the proposed method is superior to the Schechner's method. Several methods are used to recover a series of images of another target, and the result shows that the proposed method displays better performance in restoring object details and improving the overall visual effect (Fig. 6). Meanwhile, according to the value of enhancement measure evaluation function(Fig. 7), the proposed method has achieved the highest value representing the most significant improvement in image quality. Further experiments are carried out to verify the effectiveness of the proposed method for different objects in highly turbid water. Compared with other methods, the proposed method can better suppress scattered light. Therefore, the result images by the proposed method are more evenly illuminated, and the details are clearer (Fig. 8).

In this paper, in order to obtain clear images, a turbid underwater polarization image enhancement method is proposed based on frequency domain processing and polarization preservation, and the difference between the scattered light and the signal light in the frequency domain is utilized. By applying a high-pass filter on the orthogonal polarization image pair in the frequency domain, the backscattered light concentrated in the low-frequency component is initially separated and removed. Here, the optimal parameters of each filter in different concentrations and scenes are automatically searched to obtain more accurate target object information. Then the polarization degree is maintained to correlate more accurate orthogonal polarization image pairs. Finally, on the basis of the traditional physical model, the images of objects have been successfully recovered. This paper has carried out a series of experiments on different objects in underwater environments with different turbidity, and the results show that the proposed method based on frequency domain processing can effectively suppress the impact of scattered light on polarization imaging and highlight the signal light of objects. Compared with the traditional underwater polarization imaging methods, the proposed method can improve the uneven illumination problem in turbid water and significantly enhance the contrast and clarity of images, especially for highly turbid water.