Object identification from image pixel data is carried out through stages of processing in neural networks, producing characteristic- and reconstructed image data

摘要

The image data is sent to a first neural network (14) and the processor outputs one of a corresponding number (m) of characteristic data layers. This data is sent to a second neural net (18). The processors produce a layer of reconstructed image data. Weightings of neural nets (14, 18) are selected through training, such that the reconstructed image data largely corresponds with the image data at the input. An Independent claim is included for corresponding equipment.