On Comparing Two Scenarios for Probabilistic Image Modelling

Traditional scenario of probabilistic modelling is directed at generating samples having a given probability distribution. We argue that this scenario is impracticable for image modelling and should be replaced by more reasonable alternative one. This latter tends to generate samples having probabilities in the vicinity of the probability of a given training sample. Gibbs image models with multiple pairwise pixel interactions allow to explicitly implement the alternative scenario. 1 The University of Auckland, Computer Science Department, CITR, Tamaki Campus (Building 731), Glen Innes, Auckland, New Zealand 

• Type: Conference
• Year: 1998
• Where: IVCNZ
• Authors: Georgy L. Gimel'farb
• Keywords: Computer Vision, IVCNZ 1998, Samples Having Probabilities, Scenario, Traditional Scenario