Principal Warps: Thin-Plate Splines and the Decomposition of Deformations
IEEE Transactions on Pattern Analysis and Machine Intelligence
Numerical recipes in C (2nd ed.): the art of scientific computing
Numerical recipes in C (2nd ed.): the art of scientific computing
Statistical grey-level models for object location and identification
BMVC '95 Proceedings of the 6th British conference on Machine vision (Vol. 2)
Deformable template models: a review
Signal Processing - Special issue on deformable models and techniques for image and signal processing
Digital Image Processing
ECCV '98 Proceedings of the 5th European Conference on Computer Vision-Volume II - Volume II
Skeletal growth estimation using radiographic image processing and analysis
IEEE Transactions on Information Technology in Biomedicine
Pattern Recognition
Multiresolution approach to biomedical image segmentation with statistical models of appearance
Scale Space'03 Proceedings of the 4th international conference on Scale space methods in computer vision
Hi-index | 0.00 |
This paper describes a general method for segmenting articulated structures. The method is based on statistical parametral models, obtained by principal component analysis (PCA). The models, which describe shape, appearance, and topology of anatomic structures, are incorporated in a two-level hierarchical scheme. Shape and appearance models, describing plausible variations of shapes and appearances of individual structures, form the lower level, while the topological model, describing plausible topological variations of the articulated structure, forms the upper level. This novel scheme is actually a hierarchical PCA as the topological model is generated by the PCA of the parameters obtained at the lower level. In the segmentation process, we seek the configuration of the model instances that best matches the given image. For this purpose we introduce coarse and fine matching strategies for minimizing an energy function, which is a sum of a match measure and deformation energies of topology, shape, and appearance. The proposed method was evaluated on 36 X-ray images of cervical vertebrae by a leave-one-out test. The results show that the method well describes the anatomical variations of the cervical vertebrae, which confirms the feasibility of the proposed modeling and segmentation strategies.