On-line retrainable neural networks: Improving the performance of neural networks in image analysis problems

A novel approach is presented in this paper for improving the performance o f neural-network classifiers in image recognition, segmentation, or coding applications, based on a retraining procedure at the user level. The proced ure includes: 1) a training algorithm for adapting the network weights to t he current condition; 2) a maximum a posteriori (MAP) estimation procedure for optimally selecting the most representative data of the current environ ment as retraining data; and 3) a decision mechanism for determining when n etwork retraining should be activated. The training algorithm takes into co nsideration both the former and the current network knowledge in order to a chieve good generalization. The MAP estimation procedure models the network output as a Markov random field (MRF) and optimally selects the set of tra ining inputs and corresponding desired outputs, Results are presented which illustrate the theoretical developments as well as the performance of the proposed approach in real-life experiments.