Target–background contrast enhancement based on a multi-channel polarization distance model

Abstract

Abstract Stomatopods are creatures that have a unique ability to manipulate their environment by detecting polarized light for finding prey, choosing habitat, and navigation. In this study, based on the concept of polarization distance proposed by Martin J et al 2014 [Proc. R. Soc. B 281, 20131632], we have analyzed several multi-channel polarization distance models. The simulation and experimental results revealed that compared to other models, a four-channel polarization distance model can significantly enhance the contrast between the target and the background, and it exhibits excellent performance in terms of scene discrimination capability and robustness to noise. The structure and signal processing method of this model are inspired by biological polarization vision such as that of mantis shrimps. According to this method, a polarization-vision neural network is simulated with four-orientation receptor information as the input, and the network connections are realized in a cascaded order. The target–background contrast enhancement method based on this model has wide application prospects in the field of camouflage removal and target detection.