The polarization of light carries much useful information about the environment. Biological studies have shown that some animal species use polarization information for navigation and other purposes. It has been previously shown that a bio-inspired Polarization Difference Imaging technique can facilitate detection and feature extraction of targets in scattering media. It has also been established by S. Tyo1 that "Polarization Sum" and "Polarization Difference" are the optimum pair of linear combinations of images taken through two orthogonally oriented linear polarizers of a scene having a uniform distribution of polarization directions. However, in many real environments the scene has a non-uniform distribution of polarization directions. Using principal component analysis of the polarization statistics of the scene, here we develop a method to determine the two optimum information channels with unequal weighting coefficients that can be formed as linear combinations of the images of a scene taken through a pair of linear polarizers not constrained to the horizontal and vertical directions of the scene We determine the optimal orientations of linear polarization filters that enhance separation of a target from the background, where the target is defined as an area with distinct polarization characteristics as compared to the background. Experimental results confirm that in most situations adaptive polarization difference imaging outperforms "conventional" polarization difference imaging with fixed channels.