TR#175: Disparity-Space Images and Large Occlusion Stereo

Rosalind W. Picard and Alex P. Pentland

Article available in:
Proc. SPIE Conf. on
Intell. Robots and Computer Vision IX
Boston, MA
November 1991, pp. 15-26.

Markov/Gibbs random fields have been used for a variety of computer vision and image processing problems. Many of these problems are then solved using a simulated annealing type of method which involves the varying of the ``temperature,'' a scale parameter for the model. We describe the effect of varying this temperature on the resulting random field texture patterns. The idea of ``critical temperature'' has immense importance in physical pattern formation; it marks the point(s) of phase transition. In analogy to the critical temperature for an infinite system, we introduce the idea of transition temperatures for a finite texture. We show how these temperatures relate to discontinuities in the ``mixing behavior'' of the Markov/Gibbs texture. By recognizing these points of transition one can better control the textures formed by these models.

PDF . Full list of tech reports