多尺度水下偏振成像方法

水下偏振成像技术利用散射光偏振特性能够有效提高水下成像质量,在水下目标探测和识别领域具有重要应用价值.针对该技术在背景散射光和目标信息光分离时由于噪声放大现象导致重建图像质量受限的问题,提出多尺度水下偏振成像方法.该方法利用图像分层处理思想,结合小波变换的多尺度特性,对体现图像高对比度的基础层和低对比度但细节信息丰富的细节层分别进行处理,重建高对比度、高信噪比的清晰场景图像.实验结果表明,多尺度水下偏振成像方法不仅能够大幅提高对比度,复原图像细节信息,而且能够有效抑制放大噪声,提高重建图像的信噪比,在水下偏振成像领域具有良好应用前景.

Multi-scale analysis method of underwater polarization imaging

Underwater polarization imaging is a valuable technology for underwater detection and exploration, since it can provide abundant information about target scene via the removal of background light from raw images. However, in a conventional polarization imaging method, the reconstructed image has limited quality caused by the inaccurate estimation of degree of polarization (DoP) and noise amplification, which finally leads to the incomplete removal of background light. The situation becomes worse if the target and background light reach an almost equal DoP.
To date, various approaches including acoustic imaging, photoacoustic imaging, and polarization imaging have been implemented to realize underwater imaging. Notably, underwater polarization imaging is of particular interest due to its simple system structure, low cost and excellent performance in recovering target information. It mainly involves the separation of the backscattered light denoted as background light from the target scattered light acting as the target light. Removal of the background light from the raw image gives rise to a clear target image, which has been the focus of polarization imaging for a long period. The most representative approach was presented by SchechnerSchechner Y Y, Karpel N 2005 IEEE Journal of Oceanic Engineering 30 570] who utilized the DoP of background light and target light to recover clear image. Further optimization of the approach was also conducted by researchers including SchechnerTali T, Schechner Y Y 2009 IEEE Transactions on Pattern Analysis and Machine Intelligence 31 385], HuangHuang B J, Liu T G, Hu H F, Han J H, Yu M X 2016 Optics Express 24 9826], et al. However, the influence of noise amplification in the process on the reconstruction results has always been ignored, which accounts for the results to some extent though the explanation is unsatisfactory.
In this paper, we present a multi-scale polarization imaging strategy to suppress the noise amplification effect and its influence on the final results. It originates from the difference in polarization image between two diverse layers. Specifically, the image is divided into two layers, one of which is characterized by high contrast but remarkably difference between the target and background, known as base layer B_TI; the other layer is low-contrast but contains the detailed information about the target, known as detail layer D_TI. Special processes are applied to the two layers according to their characteristics, respectively. For the base layer B_TI, combined bilateral filtering is used to suppress noise. As for the detail layer, it is first processed by wavelet transform with considering its multi-resolution characteristic. After the wavelet coefficient correction via adjusting the kernel function w(x, f), the details in target image is perfected with keeping iterations. During the updating procedure, the image noise can be further suppressed. Underwater experiments are conducted in the laboratory to demonstrate the validity of the proposed method. Besides, quantitative analyses also verify the improvement in final target image.
Compared with conventional underwater polarization imaging methods, the proposed method is good at dealing with various target conditions, since it handles noise amplification without requiring any additional equipment. Furthermore, the proposed method is easy to incorporate in a conventional polarization imaging system to achieve underwater images with better quality and valid detail information. Therefore, the proposed method has more potential applications in underwater imaging.