Tasting families of features for image classification

Using multiple families of image features is a very efficient strategy to improve performance in object detection or recognition. However, such a strategy induces multiple challenges for machine learning methods, both from a computational and a statistical perspective.