Optical imaging in turbid water always suffers from poor image quality, and polarimetric imaging is an effective way for realizing clear imaging in an underwater environment. However, the traditional underwater polarimetric imaging methods can have unsatisfactory performance in the case of nonuniform optical fields because certain polarization parameters are assumed to be spatially invariant.

The authors in this paper propose an improved method of polarimetric underwater image recovery combing a deep learning technique with an unsupervised network, which can better estimate the polarization parameters to achieve clear imaging in turbid water. In particular, the background light and the degree of polarization of backscatter are considered as spatially variable parameters, and thus this method is competent for the condition of nonuniform optical fields in underwater imaging, which is the usual case of underwater imaging with active illumination. In addition, the method proposed in this paper has a lower cost in preparing the training dataset for deep learning thanks to employing an unsupervised neutral network.

---

---

 Add Comment

You must log in to add comments.