A two-layer multi-indicator framework for underwater image restoration with constrained target polarization degree

Abstract

Polarization imaging has enormous potential in underwater application scenarios. Previous polarization imaging methods mainly focused on the degree of polarization (DOP) of the backscattered light, and assumed it as a constant, resulting in no physical meaning of the optimization calculation. In this paper, particular attention is paid to the DOP of the target light, and the DOP of the backscattered light is evaluated as a function of the DOP of the target light. The global distribution of the backscattered light and its polarization degree is obtained using low-pass filtering in frequency domain. And, a novel constraint model is proposed to further confine the value range of the DOP of target light. In the proposed method, the DOP of the target light and the cutoff frequency of the low-pass filtering are chosen as the inverse parameters. To obtain the above two parameters, a two-layer multi-indicator framework is built. The proposed method can automatically realize underwater image recovery, and no background area or any prior information is required. The laboratory simulation experimental results show that the contrast and clarity of the reconstructed images have been significantly improved with different polarization characteristics under various turbidities.