Estimation of parameters of feed-back pulse coupled neural networks (FBPCNN) for purposes of microstructure images segmentation

Łukasz Łukasik, Łukasz Rauch

Faculty of Metals Engineering and Industrial Computer Science,AGH University of Science and Technology, Kraków, Poland.

DOI:

https://doi.org/10.7494/cmms.2010.1.0270

Abstract:

Although the Pulse Coupled Neural Network (PCNN) as well as FBPCNN (with feed-back) have been, since 1990, well known image analysis methods, they are still developed to solve the problems related to estimation of initial network parameters. Most of such parameters vary dependently on the character of input images (e.g. range of colors, noise strength, shapes diversity) to offer the best results. This work aims to establish parameters of the network based on FBPCNN architecture, applied in processing of images of metals’ microstructures. The paper contains detailed description of implemented neural network followed by sensitivity analysis of the network on parameters’ change. On the basis of the performed analysis, the parameters with major influence on the final results were determined and investigated in details. The results obtained in the process of image analysis by using proposed FBPCNN were passed as input data initiating Watershed algorithm for the purposes of segmentation. Results of segmentation are presented in the paper as well.

Cite as:

Łukasik, Ł., & Rauch, Ł. (2010). Estimation of parameters of feed-back pulse coupled neural networks (FBPCNN) for purposes of microstructure images segmentation. Computer Methods in Materials Science, 10(1), 8 – 15. https://doi.org/10.7494/cmms.2010.1.0270

Article (PDF):