Biomedical image segmentation

Segmentation of biomedical images separates scenes into their components based on recognition of locally similar patterns of intensity, color, texture or other features, with or without use of a priori knowledge regarding the objects or “camera” used to acquire the images. Segmented images are required for most types of object models, labeling, morphometry and geometrical investigations on imaged structures. Segmentation encompasses many methods-manual, semi-automatic and fully automatic-which are practical and useful in certain applications, as no general solution has emerged. The performance of segmentation methods is judged by comparison to manual methods, independent knowledge of truth, reproducibility and subjective criteria. Many of the methods used for segmentation can be interpreted as special cases of Grenander´s (1993, 1997) global pattern analysis, a theoretical framework for the representation of biological shape and its variability