A PCA-based approach to the three-dimensional reconstruction of human body from single frontal-view silhouette

We introduce a data-driven approach to build a human body model from a single photograph by performing Principal Components Analysis (PCA) on a database of body segments. Our work consists of two main parts, first consistent parameterization of the Civilian American and European Surface Anthropometry Resources (CAESAR) database and second segmented reconstructions of 3D bodies out of 2D photographs. Analysis on a dataset of 3D scanned surfaces has presented problems because of incompleteness on the surfaces and because of variances in shape, size and pose. In this thesis, a high-resolution generic model is aligned to data in the CAESAR database in order to obtain a consistent parameterization. A Radial Basis Function (RBF) network is built for rough deformation by using landmark information from the generic model, anatomical landmarks provided by CAESAR dataset and virtual landmarks created automatically for geometric deformation. Fine mapping then successfully applies a weighted sum of errors on both surface data and the smoothness of deformation. Compared with conventional methods, our approach has a higher efficiency in making robust alignment. Principal components analyses on segmented bodies display a richer variation than that of the whole body. Our 3D reconstruction process is composed of training and testing processes. The training aims at finding a relationship between PCA spaces in 3D and 2D. Testing process then applies this relationship to the 2D PCA-space coordinate of a new image and calculates its coordinate in the 3D PCA-space. Building several PCAs for different body parts, then making separate reconstructions and merging reconstructed body parts create the final results. We prove from practical experiments that 3D segmented body reconstructions have superiority over whole-body reconstructions. We demonstrate our approach by constructing models of a Caucasian male, an Asian male and a toddler from corresponding photographs based on a Caucasian adult oriented database. Our numerical validation includes 3D reconstructions from silhouettes of Asian males and females, Caucasian males and females, and African males and females, for which we already have their real 3D body information available from the CAESAR database. Keywords: deformation, non-rigid registration, CAESAR, RBF, PCA, segmentatio