ABSTRACT

Given a corpus of 3D laser scan data (geometry and luminance) of head and shoulder region of a subject with preplaced circular labels that mark the location of facial landmarks, we have developed a novel approach to automatically locate the preplaced markers. Accurate landmark location then allows for morphometric modeling and bio statistical analysis of facial form. We achieve this by proposing a topology independent shape matching algorithm that allows for deformable/nonrigid alignment of a template based on a modified ICP discrepancy measure. The measure involves weighted geometry and luminance components, so that markers are aligned when present, and geometrically-based landmark definition is used when markers are unavailable. The algorithm allows for per feature variance to accommodate local tolerances. We ran our algorithm on a sample of 110 scans (92 males and 18 females) with https://www.w3.org/1998/Math/MathML">94.2%https://s3-euw1-ap-pe-df-pch-content-public-p.s3.eu-west-1.amazonaws.com/9780429061943/2c936423-cf77-45da-a819-2deb9400929a/content/eq204.tif" xmlns:xlink="https://www.w3.org/1999/xlink"/> success overall with approximately less than 23 148seconds for 24 landmarks search including I/O time. We plan to extend the algorithm to include a global warp from the model to the target mesh and thus creating a topology and pre-processing free method to establish a dense correspondence between two meshes which will have significant impact in the area of shape matching.

Model pointsets drawn superimposed over an input pointset; (a) an input model with the local errors for a single landmark highlighted in red (in the callout); (b) The same mesh while another landmark is being aligned, viewed from the left; (c) examples of a head stabilizer; (d) a pointset with a fair amount of scanning artifacts, landmarks are successfully identified in green. https://s3-euw1-ap-pe-df-pch-content-public-p.s3.eu-west-1.amazonaws.com/9780429061943/2c936423-cf77-45da-a819-2deb9400929a/content/fig15_1.jpg" xmlns:xlink="https://www.w3.org/1999/xlink"/>