A Survey of Flat Graphical Model in Image Understanding

Flat Graph Model has become a very active direction in Image Understanding (IU) field, which constructs the probabilistic model of analysis object, processes parameter learning and probability inference, and obtains the final recognition result by analyzing maximum a posteriori. Image Under-standing could be regarded as labeling each pixel or patch independently. Flat graph model, latent generative model, and its applications in the task of IU are investigated in this paper. First of all, latent "topics" are discovered by using bag-of-words model in detail, and the generative model from the statistical text literature here is applied to a bag of visual words representation for each image. Subsequently, probabilistic learning and inference are processed on the topic distribution vector for each image. At last, the application and prospect of flat graph model are briefly explored.