Abstract
This paper proposes a novel fast boundary tracing scheme which naturally encodes the prior knowledge. Another advantage is that this scheme is independent of the subsequent segmentation process. Therefore, it can serve as a pre-processing step in any segmentation scheme. For example, it could be used to automatically generate a close-to-boundary initialization for all types of deformable models. Topographic Independent Component Analysis (TICA) based feature exaction technique is adopted for learning a representation from a set of image patches in an unsupervised way. During learning, a topographic map of basis components emerge. A unique intelligent contour generation procedure is also proposed. Experimental results on abdominal CT images demonstrate the potential of our approach.