The standard approach to image reconstruction using deep learning is to use clean image priors for training purposes. In this project, we attempt to achieve denoising without using a clean image prior and yet, achieving a performance comparable to, or sometimes, even better than that obtained using the conventional approach.