Shape and Appearance Based Window Extraction

We present a method to extract windows in facade images based on shape and appearance model. We break the extraction problem down into two separate stages, the data fitting stage in which we find a model that best fits observed image and prior information, and the subsequent classification stage in which we use the model found to make class assignment. We formulate the task as a constrained optimization problem and use an iterative procedure to find optimal value. We decrease computational complexity of the method by reducing the size and dimension of the model parameter space. Both shape and appearance model are treated in a single framework and allow extraction of a variety of common window types.