Learning Generic Prior Models for Visual Computation

This paper presents a novel theory f o r learning generic prior models f r o m a set of observed natural images based o n a minimax entropy theory that the authors studied in modeling textures. W e start by studying the statistics of natural images including the scale invariant properties, then generic prior models were learnt t o duplicate the observed statistics. The learned Gibbs distributions confirm and improve the forms of existing prior models. More interestingly inverted potentials are found to be necessary, and such potentials f o r m patterns and enhance preferred image features. The learned model is compared with existing prior models in experiments of image restoration.