Learning to Parse Images
Geoffrey E. Hinton, Zoubin Ghahramani, Yee Whye Teh · 1999
We describe a class of probabilistic models that we call credibility networks. Using parse trees as internal representations of images, credibility networks are able to perform segmentation and recognition simultaneously, removing the need for ad hoc segmentation heuristics. Promising results in the problem of segmenting handwritten digits were obtained. 1 Introduction The task of recognition has been the main focus of attention of statistical pattern recognition for the past 40 years. The paradigm problem is to classify an object from a vector of features extracted from the image. With the advent of backpropagation [1], the choice of features and the choice of weights to put on these features became part of a single, overall optimization and impressive performance was obtained for restricted but important tasks such as handwritten character identification [2]. A significant weakness of many current recognition systems is their reliance on a separate preprocessing stage that se...