 |
|
|
| project results: |
Hierarchical Conditional Random Field for Multi-class Image ClassificationMichael Ying Yang, Martin Drauschke, Wolfgang FörstnerDepartment of Photogrammetry University of Bonn October, 2009 Classification of image components in meaningful categories is a challenging task due to the ambiguities inherent into visual data. On the other hand, image data exhibit strong contextual dependencies in the form of spatial interactions among components. Conditional random fields (CRFs) have been proposed as a principled approach to modeling the interactions between labels in such problems using the tools of graphical models. A conditional random field is a model that assigns a joint probability distribution over labels conditional on the input, where the distribution respects the independence relations encoded in a graph. Standard CRFs act at a very local level and represent a single view on the data typically represented with unary and pairwise potentials. In order to overcome those local restrictions, we introduce multiple layers with associated unary potentials and pairwise potentials, to enhance the model by evidence aggregation on local to global level. We propose a method that explicitly models region adjacent neighborhood information within each layer and region hierarchical information between the layers, using global image features as well as local ones for observations in the model. We denote this model as Hierarchical Conditional Random Fields (HCRF).
References
|
