3D HEAD POSE NORMALIZATION WITH FACE GEOMETRY ANALYSIS, GENETIC ALGORITHMS AND PCA

摘要: In this paper, a software toolchain is presented for the fully automatic alignment of 3D human face model. Beginning from point cloud head (previously segmented its background), pose normalization obtained using an innovative and purely geometrical approach. order to solve six degrees freedom raised by problem, we first exploit face's natural mirror symmetry; secondly, analyze frontal profile shape; finally, align model's bounding box according position tip nose. The whole procedure considered as two-fold, multivariable optimization problem which addressed use multi-level, genetic algorithms greedy search stage, with latter being compared against standard PCA. Experiments were conducted utilizing GavabDB database took into account proper preprocessing stages noise filtering model reconstruction. Outcome results reveal strong validity in approach, however, at price high computational complexity.