Abstract: The traditional independent component analysis does not have the anti-interference ability, and it needs to make the noisy image blind separation under the condition that the number of sources is known. To solve this problem, a blind separation method is proposed for noisy images based on variational Bayesian independent component analysis (VbICA). Contrast with the conventional separation methods, the proposed algorithm can directly and effectively separate the noisy images and has strong anti-interference ability. In addition, the algorithm estimates the number of sources based on the reliability of different models. The experiment results show that this algorithm is very effective.