| ECVision indexed and annotated bibliography of cognitive computer vision publications This bibliography was created by Hilary Buxton and Benoit Gaillard, University of Sussex, as part of ECVision Specific Action 8-1 The complete text version of this BibTeX file is available here: ECVision_bibliography.bib
D. De Ridder and R. P. W Duin and P. W. Verbeek and L. J. {Van Vliet}
ABSTRACT
In this paper the implementation of non-linear image processing operations using feed-forward neural networks is studies. As an example, the Kuwahara filter for edge-preserving smoothing is chosen, an interesting algorithm since it is highly non-linear and inherently modular. Several neural networks were trained, designed both with and without incorporating prior knowledge about the problem. It will be shown that the mean squared error (MSE) criterion used in training the networks does not give a good indication of filter performance. As an alternative, measures are proposed which can capture the dual goals of edge sharpening and region smoothing in Kuwahara filtering. Using these, it is shown that the use of prior knowledge in this application is a prerequisite for good performance. |
|
| Site generated on Friday, 06 January 2006
|