Noise modelling using Deep CNN for Terahertz Super-Resolution Imaging

This work focuses on a generalized noise model for terahertz imaging. Statistical noise masks are constructed to analyze Gaussian, quantization, data-drop and speckle noises in terahertz images, and are analyzed using the Noise2Clean algorithm. The consensus of the Noise2Noise algorithm is also analyzed to obtain super-resolution with only noisy data. The results show that a deep convolution network can be trained using numerical probability distribution models for denoising terahertz images with only noisy training samples.