Mixture of Matrix Normal Distributions for Color Image Inpainting

Gaussian mixture model is commonly used as image prior model to solve image restoration problem. However, vector representation leads to lose the inherent spatial relevant information and cause unstable estimation. In this paper, a mixture of matrix normal distributions (MMND) based image restoration algorithm is proposed, which incorporates the hidden structural information into prior image modeling. MMND is used as the prior image model and expectation maximization algorithm is used to optimize the maximum posterior criterion. Experiments conducted on color images indicate that MMND can achieve better peak signal to noise ratio (PSNR) as compared to other state-of-the-art methods.