Graph Cut Based Segmentatioln Of Multimodal Images

This work proposes a new semi-unsupervised Maximum- A-Posteriori (MAP) based segmentation framework of multimodal images. In this work a joint Markov Gibbs random field (MGRF) model is used to describe the image. However, the main focus here is a more accurate model identification. We propose a new analytical approach to estimate spatial interaction potentials for the MGRF model. For a known number of classes in the given image, the empirical distributions of this image signals are precisely approximated by a linear combination of Gaussian (LCG) distributions with positive and negative components. The proposed framework consists of three stages. The first stage is the image signal modelling, and initial labeling stage. In the second stage the new analytically estimated potential is used to identify the spatial interaction between the neighboring pixels. Finally, an energy function using the previous models is formulated, and is globally minimized using graph cuts. Experimental results show that the developed technique gives promising accurate results compared to other known algorithms.