Inexact graph matching using stochastic optimization techniques for facial feature recognition

We propose a formalization of model-based facial feature recognition as an inexact graph matching problem, one graph representing a model of a face and the other an image where recognition has to be performed. The graphs are built from regions and relationships between regions. Both nodes and edges are attributed. A global dissimilarity function is defined based on comparison of attributes of the two graphs, and accounting for the fact that several image regions can correspond to the same model region. This junction is then minimized using several stochastic algorithms.