Unsupervised Image Segmentation Using A Simple MRF Model with A New Implementation Scheme

Huawu Deng, David A. Clausi

Abstract: A Markov random field (MRF) model with a new implementation scheme is proposed for unsupervised image segmentation based on image features. The traditional two-component MRF model for segmentation requires training data to estimate necessary model parameters and is thus unsuitable for unsupervised segmentation. The new MRF model overcomes this problem by introducing a function-based weighting parameter between the two components. This new MRF model is able to automatically estimate model parameters and is demonstrated to produce more accurate image segmentations than the traditional model using a variety of imagery.