A Novel Method for Scene Categorization with Constraint Mechanism Based on Gaussian Statistical Model
Jun Gao · Dianzi xuebao · 2009
Focusing on scene categorization,we presented a method based on Gaussian statistic model.Without feature hand-annotation for scene category,we can use unsupervised clustering and parameter estimation in Gaussian model to build corresponding probabilistic relations between scene labelling semantics and textons information.The experiments reveal that the proposed method performs more efficiently than other supervised ones and has the quite approximate results with complex hierarchical models.At the same time,we also studied constraint mechanism for object analysis primarily.Finally,we can draw conclusions that scene information provide efficient prior knowledge for object analysis by experiments.With constraints on possibilities of the object appearance in different scenes,object accuracy can be improved.