The present invention relates to the field of digital image processing and computer vision, and in particular to a Bayesian image denoising method based on distribution constraints of noisy images. In Bayesian posterior probability theory, estimating the noise-free images from the noisy images depends on modeling the previous distribution of the noise-free images. The present invention first proposes a method for obtaining the distribution of noise-free images by learning from noisy image samples, knowing the additive noise distribution model. The present method transforms the constraint of the prior distribution of the noise-free images into the constraint of the prior distribution of the noisy images for Bayesian denoising. The present invention further proposes a Bayesian denoising implementation method for unsupervised training of a picture denoising neural network. The present method makes it possible to fully utilize image samples containing noise in order to accurately learn the implicit distribution properties of noise-free images, thereby realizing efficient image denoising.
展开▼