The quality of underwater polarization imaging is mainly affected by the polarization properties of the target and impurity particles. Traditional methods often assume uniform polarization characteristics of the target, which make it difficult to address the restoration issues of complex targets. In response, a partition-based method for recovering underwater polarization images is proposed. The method involves pre-processing the image using the Gaussian curvature filtering algorithm, partitioning the image based on polarization information. In addition, a joint image evaluation method is used to achieve restoration of complex polarized characteristic targets. The method estimates the value of the reflected light polarization of one partition to estimate the value of the next partition and links the polarization values of each partition. Our approach achieves clear restoration results for multiple targets or complex structural objects underwater. Achieving significant improvement in image quality in multi-target underwater scenes, our method is highly effective for complex underwater environments. Experimental results show that our method, when compared with three other newer methods on multiple images of different targets and under varying scattering conditions, achieves an average increase of 617% in the standard deviation image evaluation index and a 61% optimization in the natural image quality evaluator index. Furthermore, our method is robust for different degrees of water turbidity.