Polarization difference imaging technique can effectively solve the underwater image deterioration problem that is caused by the interaction between light and water. Therefore, it has a significant application value in detecting and recognizing underwater target. In a traditional polarization difference imaging system, the object image is carried out by the common-mode rejection of background scattering light. However, the polarization state of the background scattering light is unknown, so the polarization difference imaging is realized by the irregular mechanical rotation of the optical polarization analyzer with two orthogonal polarization orientations. Therefore, it needs more time to determine the optimum detection angle of the polarization analyzer and cannot perform real-time underwater imaging, which restricts the rapid detecting function in the process of underwater imaging. In this paper, the detection principle of underwater polarization difference imaging is considered to exploit the difference in the polarization angle between background scattering light and target light. According to Marius's law, the physical model of polarization difference imaging is that the common-mode rejection of background scattering light will be achieved when the angles between the vibration direction of background and the two orthogonal polarization orientations are 45. Because the Stokes vector can be used to express the polarization angle of light, we propose the principle and construction of a computational polarization difference imaging system for real-time underwater imaging by incorporating the Stokes vector into the established model. It replaces the mechanical rotation of the polarization analyzer in a traditional polarization difference imaging system with the information processing of the Stokes vector. The experimental results show that the proposed method not only has the same effective performance as the conventional polarization difference imaging compared with the regular imaging, but also can improve the blurred imaging performance caused by an underwater scattering effect as well as increase the underwater detection distance. This method realizes rapid underwater target detection and recognition because it saves a large amount of time compared with the traditional one. Further, if we combine this method with the current polarization imaging instruments that capture the Stokes vector instantaneously, then a real-time automatic underwater polarization imaging can improve the efficiency of the underwater target detection and recognition. These findings are helpful for designing and developing the underwater polarization difference imaging systems.