Image Domain Conversion Algorithm Based on Generative Adversarial Network

In this study, we proposed Enhanced-Pix2Pix as an infrared image migration algorithm based on Pix2Pix to address the issues of loss of detail in textures, noise, and poor conversion performance in traditional generative adversarial networks during the process of transferring visible light images to infrared images. First, a generator structure consisting of a deep radial connected encoder and a feature fusion module based on a convolution block attention module, and a multi-scale discriminator and composite loss function were designed to enhance the ability of the model to discriminate between objects. Experiments were conducted on the RGB-T234 visible-infrared paired dataset. The experimental results show that the proposed image domain conversion algorithm improved the peak signal to noise ratio (PSNR) by 5.716 and 1.827 and the structural similarity index measure (SSIM) by 0.134 and 0.049, respectively, compared to the traditional generative adversarial network models CycleGAN and Pix2Pix.