Longitudinal Face Modeling via Temporal Deep Restricted Boltzmann Machines

摘要

Modeling the face aging process is a challenging task due to large andnon-linear variations present in different stages of face development. Thispaper presents a deep model approach for face age progression that canefficiently capture the non-linear aging process and automatically synthesize aseries of age-progressed faces in various age ranges. In this approach, wefirst decompose the long-term age progress into a sequence of short-termchanges and model it as a face sequence. The Temporal Deep Restricted BoltzmannMachines based age progression model together with the prototype faces are thenconstructed to learn the aging transformation between faces in the sequence. Inaddition, to enhance the wrinkles of faces in the later age ranges, the wrinklemodels are further constructed using Restricted Boltzmann Machines to capturetheir variations in different facial regions. The geometry constraints are alsotaken into account in the last step for more consistent age-progressed results.The proposed approach is evaluated using various face aging databases, i.e.FG-NET, Cross-Age Celebrity Dataset (CACD) and MORPH, and our collectedlarge-scale aging database named AginG Faces in the Wild (AGFW). In addition,when ground-truth age is not available for input image, our proposed system isable to automatically estimate the age of the input face before aging processis employed.