Multigrid MRF based picture segmentation with cellular neural networks

Abstract

Due to the large computation power needed in image processing methods based on Markovian Random Field (MRF) [6], new variations of basic MRF models are implemented. The Cellular Neural Network [5,14,15] (CNN) architecture, implemented in real VLSI circuits, is of superior speed in image processing. This very fast CNN can implement the ideas of existing MRF models. which would result in real-time processing of images. On the other hand this VLSI solution gives new tasks since the CNN has a special local architecture [4], but it is already shown that a type of MRF image segmentation with Modified Metropolis Dynamics (MMD [9]) can be well implemented in the CNN architecture [18]. In this paper, we address the improvement of the existing CNN method [17]. We have tested different multigrid models and compared segmentation results. The main reason for this research is to find proper implementation of the CNN-MRF technique on CNNs taking into consideration the abilities of today's and future's VLSI CNN systems.