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

Doulamis, AD; Doulamis, ND; Kollias, SD

dc.contributor.author	Doulamis, AD	en
dc.contributor.author	Doulamis, ND	en
dc.contributor.author	Kollias, SD	en
dc.date.accessioned	2014-03-01T01:15:45Z
dc.date.available	2014-03-01T01:15:45Z
dc.date.issued	2000	en
dc.identifier.issn	1045-9227	en
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/13716
dc.subject	image analysis	en
dc.subject	MPEG-4	en
dc.subject	neural-network retraining	en
dc.subject	segmentation	en
dc.subject	weight adaptation	en
dc.subject.classification	Computer Science, Artificial Intelligence	en
dc.subject.classification	Computer Science, Hardware & Architecture	en
dc.subject.classification	Computer Science, Theory & Methods	en
dc.subject.classification	Engineering, Electrical & Electronic	en
dc.subject.other	MODEL	en
dc.subject.other	SEGMENTATION	en
dc.subject.other	VIDEO	en
dc.subject.other	ALGORITHMS	en
dc.subject.other	FIELDS	en
dc.title	On-line retrainable neural networks: Improving the performance of neural networks in image analysis problems	en
heal.type	journalArticle	en
heal.identifier.primary	10.1109/72.822517	en
heal.identifier.secondary	http://dx.doi.org/10.1109/72.822517	en
heal.language	English	en
heal.publicationDate	2000	en
heal.abstract	A novel approach is presented in this paper for improving the performance of neural-network classifiers in image recognition, segmentation, or coding applications, based on a retraining procedure at the user level. The procedure includes: 1) a training algorithm for adapting the network weights to the current condition; 2) a maximum a posteriori (MAP) estimation procedure for optimally selecting the most representative data of the current environment as retraining data; and 3) a decision mechanism for determining when network retraining should be activated. The training algorithm takes into consideration both the former and the current network knowledge in order to achieve good generalization. The MAP estimation procedure models the network output as a Markov random field (MRF) and optimally selects the set of training 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.	en
heal.publisher	IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC	en
heal.journalName	IEEE TRANSACTIONS ON NEURAL NETWORKS	en
dc.identifier.doi	10.1109/72.822517	en
dc.identifier.isi	ISI:000085524000014	en
dc.identifier.volume	11	en
dc.identifier.issue	1	en
dc.identifier.spage	137	en
dc.identifier.epage	155	en