An automated volumetric segmentation system combining multiscale and statistical reasoning

An automated volumetric image segmentation algorithm is proposed. This method is fast and unsupervised, automatically estimating required parameters including optimal segment number selection using Bayesian inference. In the wavelet domain, Gaussian mixture modeling (GMM) is used to achieve a baseline scene estimate. This estimate is then refined to consider spatial correlations using a Markov random field model (MRFM). The application of this system to three-dimensional biomedical image volumes is discussed. This approach delivers promising results in terms of the identification of inherent image features.