Light field depth estimation method based on unsupervised deep learning

The invention discloses a light field depth estimation method based on unsupervised deep learning. According to the invention, an unsupervised loss function is designed; meanwhile, a group of sub-aperture images arranged in a 3 * 3 mode are extracted from the light field image to serve as input of a light field depth estimation network, a disparity map of the central sub-aperture image is output,and end-to-end training is achieved. The method comprises the steps of S1, preparing a light field data set, and making a training set and a test set; S2, building an unsupervised light field depth estimation network; S3, designing an unsupervised light field depth estimation loss function; S4, training an unsupervised light field depth estimation network by using the training set. According to the invention, by using the self-built network structure and the loss function, the precision superior to that of other unsupervised depth estimation methods can be obtained on a 4D light field data setevaluation website provid