Using structured input patterns for neural based image recognition

Describe an application of neural networks to the recognition of faces. A parallel implementation of a learning algorithm is presented, and the authors discuss some critical aspects of face recognition and describes the goals of the system. The architecture of the system is presented; both the overall hierarchical structure and the function and structure of the different modules are described in some detail. Particular attention is paid to the concept of meta-neural net, that is the use of a neural net which 'learns the behaviour of others neural nets, which is introduced for enhancing flexibility and efficiency of the system. Some experimental results are given, which show that the architecture shows a high generalization capability and allows the recognition of images with different kind of noise better than each single network can do. Finally a possible parallel implementation of the system with a preliminary evaluation of its performances is presented.<
>