A method for mixed states texture segmentation with simultaneous parameter estimation

作者：Mailing A^*; Cernuschi Frias B

来源：Pattern Recognition Letters, 2011, 32(15): 1982-1989.

DOI：10.1016/j.patrec.2011.07.022

• Motion textures
• Segmentation
• Expectation Maximization
• Pseudo-likelihood
• Markov Random Fields

In this work a method for mixed-state model motion texture segmentation and parameter estimation is presented. We use the Expectation Maximization algorithm for mixture parameter estimation, introducing the Gibbs distribution for moving points, excluding zero discrete component associated with no motion regions. We use then the a posteriori probabilities to generate an alternative field to segment the textures according to its statistical parameters.

• 出版日期2011-11-1