Image compression with a hierarchical neural network

A neural network data compression method is presented. This network accepts a large amount of image or text data, compresses it for storage or transmission, and subsequently restores it when desired. A new training method, referred to as the Nested Training Algorithm (NTA), that reduces the training time considerably is presented. Analytical results are provided for the specification of the optimal learning rates and the size of the training data for a given image of specified dimensions. Performance of the network has been evaluated using both synthetic and real-world data. It is shown that the developed architecture and training algorithm provide high compression ratio and low distortion while maintaining the ability to generalize, and is very robust as well.