Terahertz Image Restoration with “Zero-Shot” Super-Resolution

The spatial resolution of terahertz (THz) image is often degraded by many factors during the imaging process. Therefore, a “Zero-Shot” super-resolution CNN framework, which is trained only by the input image itself, is proposed to cope with those degradation factors. With this unsupervised framework, the restoration level of the CNN can be flexibly adjusted based on the input THz image in order to achieve the best restoration effect. Experimental results on both the simulated data and real tested THz data have proved the effectiveness of our method.