Infrared and Visible Image Fusion Based on Biological Vision

Abstract

Infrared images can acquire salient targets, while visible images contain richer details. It is vital to fuse these two types of images. Benefiting from the existence of the dual-mode cellular mechanism, the rattlesnake is able to process and fusion infrared and visible signals, improving the predatory ability. In this paper, we design an auto-encoder fusion network based on the visual adversarial receptor domain. In this network, we build a feature-level fusion strategy based on the dual-modal cell mechanism which is simulated by the human visual cell’s center-antagonistic receptor domain. Meanwhile, we optimize the feature extraction and feature reconstruction modules in fusion network. By realized the combined research of biological vision and computer vision, our network delivers a better performance than the state-of-the-art methods in both subjective and objective evaluation.