A typical image de-blurring algorithm consists of a blurring model, a prior distribution on natural images, and a noise model. Existing algorithms dont work well on JPEG images, since they all assume the noise is independently identically distributed on every pixel, which works well on raw pixel data but fails to capture the quantization loss introduced through JPEG compression. Thus, Ill be working on a new noise model that explicitly takes into account the JPEG uncertainty which varies based on the frequency in the Discrete Cosine Transform (DCT) domain, in order to more effectively de-blur JPEG images.