The automatic recognition of digital biomedical images is a very high level complex task, involving both expertise about image processing tools and domain dependent specific knowledge. In order to face this complexity, in late 80's have been proposed the so-called Knowledge Based Vision Systems; KBVS system where a set of production rules was used to explicitly represent the expert's experience. In spite of the formal correctness if this approach, the KBVS don't seem to have, at present, a wide utilization in real applications. The main failure of the symbolic approach is probably due to the difficult of fully elicit the knowledge of a skilled analyst. The Artificial Neural Networks seem to be a tool to overcome this difficulty, because they are systems able to determine the relationship between the input to and the output from the network, directly from the examples used in the training phase. In this paper is presented an approach based on the concept of multi-modular neural architecture, obtained by combining small specialized networks.
Un approccio a reti neurali per sistemi di immagini basati sulla conoscenza
PASQUARIELLO G
1994
Abstract
The automatic recognition of digital biomedical images is a very high level complex task, involving both expertise about image processing tools and domain dependent specific knowledge. In order to face this complexity, in late 80's have been proposed the so-called Knowledge Based Vision Systems; KBVS system where a set of production rules was used to explicitly represent the expert's experience. In spite of the formal correctness if this approach, the KBVS don't seem to have, at present, a wide utilization in real applications. The main failure of the symbolic approach is probably due to the difficult of fully elicit the knowledge of a skilled analyst. The Artificial Neural Networks seem to be a tool to overcome this difficulty, because they are systems able to determine the relationship between the input to and the output from the network, directly from the examples used in the training phase. In this paper is presented an approach based on the concept of multi-modular neural architecture, obtained by combining small specialized networks.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
