Combining model-based classifiers for face localization

We present a method to localize a face in a color image combining connexionist models (auto-associator networks), an ellipse model based on the orientation of the gradient and a skin color model. A linear combination of each model response is performed. Given an input image, we compute a kind of probability map on it with a sliding window. The face position is then determined as the location of the absolute maximum over this map. Improvement of localization rates of individual detectors is clearly shown and results are very encouraging.