Diorthosis prediction based on landmarks detected by layered diffusion and contours deformed from template

This paper proposes a new scheme to locate cephalogram landmarks and predict diorthosis surgery results. It applies multi-scale diffusion for feature point detection and template warping for structure fitting. The localization makes uses of homogeneity inside a region and saltations across edges, it includes three steps: firstly, layered anisotropic diffusions are carried out by an adaptive nonlocal filter with its size adjusted for facial outline, large and tiny bone structures respectively. Secondly, feature points are determined on edge maps of layered diffusions by comparison with a knowledge-based template. Thirdly, landmarks of facial outline are obtained via subtraction a bone-structure image and positioning with extremum algorithm. On the prediction stage, diorthosis result is estimated using the detected datum mark for template deformation, the warping rules are based on the statistical correlation between a template shift and the corresponding changes of facial outline. Experiments show that our approach takes the advantages of layered edge preserving and easy landmark extraction, it makes a more efficient landmark location and template matching for surgery result prediction.