Mean field and information theoretic algorithms for direct segmentation of tomographic images

The weak membrane model with optimization by mean field annealing is applied to the direct segmentation of tomographic images. Models based on the minimum description length principle that include penalties for measurement error, boundary length, regions, and means are also introduced. Outliers are prevented by upper and lower bound constraints on pixel values. Several models are generalized to three-dimensional images. The superiority of the present models to convolution backprojection is demonstrated experimentally.<>