Framework for face aging modeling:

Compositional face representation + Markov dynamic model

1. Our Objectives

Simulate facial appearances changes during aging process to (i) Looking for missing children or wanted fugitives. (ii) Help improving performances of face recognition systems. (iii) Generating special effects in filmmaking and game development. (iv) Preview results of plastic surgery. .

2. Challenges

High dimension------a lot of variations occur, including cranium, muscles, skin, et al. Highly nonlinear------different features occur at different time and the aging rate is also time-variant. Intrinsically uncertain------aging is affected by gene and external factors simultaneously.

3. Our Solution

To deal with the problems of lacking long term dense aging databases to build face aging models, we propose an approach of learning long term face aging patterns from partially dense aging datasets. The approach is based on two assumptions: (i) Decompose human faces into semantic regions to reduce dimension; (ii) Decompose age range into small age spans to reduce nonlinearity. (iii) Simulate the uncertainty of face aging by probabilistic sampling.

4. Dynamic Aging Modeling

Discrete Markov process(PAMI paper) Continous Markov process((ICCV paper)

5. Discussion and Future work

Under current avaiable databases, the framework is necessary and feasible. With the development of aging datasets, we will continue to improve our models.