DCGANs for image super-resolution , denoising and debluring

Advance of computational power and big datasets brings the opportunity of using deep learning methods to do image processing. We used deep convolutional generative adversarial networks (DCGAN) to do various image processing tasks such as super-resolution, denoising and deconvolution. DCGAN allows us to use a single architecture to do different image processing tasks and achieve competitive PSNR scores. While the results of DCGAN shows slightly lower PSNR compared to traditional methods, images produced by DCGAN is more appealing when viewed by human. DCGAN can learn from big datasets and automatically add high-frequency details and features to images while traditional methods can’t. The generatordiscriminator architecture in DCGAN pushes it to generate more realistic and appealing images.