Modified Generative Adversarial Network for Super-Resolution of Terahertz Image

Terahertz (THz) images have low spatial resolution, blurring contour features and high background noise owing to the limitation of terahertz (THz) wavelengths and the THz imaging systems. We have proposed a modified Generative Adversarial Network (GAN) for super-resolution (SR) purpose. To fit the THz images, we design a kind of image degradation model to generate low-resolution images with Gaussian blur and white Gaussian noise. We establish a dataset of damage images in the field of non-destructive testing (NDT) for training and testing. The experimental results on THz images demonstrate that the improved GAN model can improve the quality of THz images effectively. Our method can be beneficial to improve the accuracy of THz NDT with low resolution.