Estimating parameters of a synthetic deformable human head model using a set of photographs

In this paper a new method for estimating the parameters of a synthetic deformable model of human head using a set of photographs is proposed. As opposed to statistical models, synthetic models have a well defined correspondence between the parameters and morphological features of the human head, though this bounds variability and complicates the task of fitting the model to photographs. A novel heuristic coarse-to-fine algorithm based on the psychophysical process of face perception and recognition by humans is proposed. A user study for evaluating the perceived quality of the generated models is also designed and conducted. Applicability of the proposed method for human-head modeling from photographs is corroborated by results of experiments on real world data.