METHOD AND DEVICE FOR TRAINING NEURAL NETWORK, IMAGE PROCESSING METHOD AND DEVICE AND STORAGE MEDIUM

摘要

Disclosed are a method and device for training a neural network, an image processing method and device and a storage medium. The neural network comprises a projection domain network for processing input projection data to obtain estimated projection data; a parsing and reconstruction network layer for obtaining a reconstructed image from the estimated projection data; an image domain network for processing the reconstructed image to obtain an estimated image; a projection layer for obtaining a projection result of the estimated image; and a statistical model layer for determining a statistical model-based consistency of the input projection date, the estimated projection data and the projection result of the estimated image. The neural network may further comprise a prior model layer. The method comprises: adjusting convolution kernel parameters of the image domain network and the projection domain network using a consistency cost function of a data model based on the input projection data, the estimated projection data and the projection result of the estimated image. By use of the described solution, the neural network obtained by training can reconstruct an image having higher quality when projection date has a defect.