Multi-View Random Fields and Street-Side Imagery

In this paper, we present a method that introduces graphical models into a multi-view scenario. We focus on a popular Random Fields concept that many researchers use to describe context in a single image and introduce a new model that can transfer context directly between matched images – Multi-View Random Fields. This method allows sharing not only visual information between images, but also contextual information for the purpose of object recognition and classification. We describe the mathematical model for this method as well as present the application for a domain of street-side image datasets. In this application, the detection of façade elements has improved by up to 20% using Multi-view Random Fields.