Methods for image recovery, processing methods for image encoding, and methods read by non-force computers

Abstract

The invention relates to the decoding and encoding of images of complementary streams based on neural networks, with the image being represented by the input set of samples obtained from the bit stream. The image is reconstructed from the output subsets which are generated as a result of processing the input set L. The input set is divided into several input subsets Li. Each of the input subsets is processed by a neural network with one or more layers. Neural networks use multiple input subsets as input and generate a single output subset. By combining the output subsets, the image is reconstructed. Specifically, the size of at least one input subset is smaller than the size required to obtain the size of the corresponding output subset after processing by one or more layers of the neural network. The term "processing" also refers to the methods and machines for reproducing from a bit stream, the methods and machines for processing, and the non-transferable storage medium.