`a`

Improvement of image processing by using homogeneous neural networks with fractional derivatives theorem

Pages: 505 - 514, Issue Special, September 2011

 Abstract        Full Text (341.5K)              

Zbigniew Gomolka - University of Rzeszow, Institute of Technology, 35-959 Rzeszow, 16A Rejtana Str., Poland (email)
Boguslaw Twarog - University of Rzeszow, Institute of Technology, 35-959 Rzeszow, 16A Rejtana Str., Poland (email)
Jacek Bartman - University of Rzeszow, Institute of Technology, 35-959 Rzeszow, 16A Rejtana Str., Poland (email)

Abstract: The present paper deals with the unique circumvention of designing feed forward neural networks in the task of the interferometry image recog- nition. In order to bring the interferometry techniques to the fore, we recall briefly that this is one of the modern techniques of restitution of three di- mensional shapes of the observed object on the basis of two dimensional flat like images registered by CCD camera. The preliminary stage of this process is conducted with ridges detection, and to solve this computational task the discussed neural network was applied. By looking for the similarities in the biological neural systems authors show the designing process of the homogeneous neural network in the task of maximums detection. The fractional derivative theorem has been involved to assume the weight distribution function as well as transfer functions. To ensure reader that the theoretical considerations are correct, the comprehensive review of experiment results with obtained two dimensional signals have been presented too.

Keywords:  Homogeneous neural network, Fractional derivative, Interferometry image
Mathematics Subject Classification:  Primary: 68T45, 94A08, 68T05; Secondary: 47B39, 68W10

Received: July 2010;      Revised: April 2011;      Published: October 2011.