A Bayesian framework for computer vision

Abstract

In many computer vision applications the input images are noisy and the data possibly sparse. These applications include, (i) identification of abnormalities in human organs, (ii) the computation of anatomical and physiological properties, both for medical images, e.g. liver tumors from CT images, human heart volume from MRI imagery, (iii) depth information for moving robots or automatic mapping of satellite image pairs, and (iv) inspection of faults on large scale manufactures. The common problems are then to “clean” the images from the noisy and/or sparse data and to segment the image. Particular emphasis is placed on image segmentation, that is, separating objects from background.