Compact Part-Based Image Representations

Learning compact, interpretable image representations is a very natural task which has not been solved satisfactorily even for simple classes of binary images. In this paper, we review various ways of composing parts (or experts) for binary data and argue that competitive forms of interaction are best suited to learn low-dimensional representations. We propose a new rule which discourages parts from learning similar structures and which penalizes opposing expert opinions strongly so that abstaining from voting becomes more attractive. Using a process of oversimplification and correction we show in experiments that very intuitive models can be obtained.