Learning displacement experts from multi-band images for face model fitting

Models are often used to gain information about real-world objects. Their parameters describe various properties of the modeled object, such as position or deformation. In order to fit the model to a given image, displacement experts serve as an update function on the model parameterization. However, building robust displacement experts is a non-trivial task, especially in real-world environments. We propose a novel approach that learns displacement experts from a multi-band image representation which is specifically tuned towards the task of face model fitting. We provide the fitting algorithm not only the original image but an image representation that reflects the location of several facial components within the face. To demonstrate its capability to work robustly not only in constrained conditions, we integrate the Labeled Faces In The Wild database, which consists of images that have been taken outside lab or office environments. Our evaluation demonstrates, that the information provided by this image representation significantly increases the accuracy of the model parameter estimation.