ABSTRACT
An object of interest in an image can be characterized to some extent by the shape of its external boundary. It is therefore important to develop procedures for boundary extraction in problems of detection, tracking, and classification of objects in images. Active contour algorithms have become an important tool in image segmentation for object detection [4, 5, 9]. As segmentation
in Bayesian
more sophisticated, they are tested in more difficult imaging environments of real-world scenarios where images do not have enough contrast to provide sharp boundaries, there is some occlusion of the target, or there exists target-like clutter or noise. Thus, it is of increasing importance that boundary extraction algorithms make use of prior knowledge about the expected target class in order to help compensate for the lack of clear data. This is accomplished by influencing the contour evolution in part with a shape prior, a statistical model derived from a set of known training shapes, in a Bayesian active contour approach [3, 7, 13, 18].
