Time dependent Markov matrices for automated image analysis

The exploitation of time dependent Markov matrices for automate image analysis is discussed. Markov process theory provides powerful techniques for automated image understanding algorithms. This paper investigates Markov chains defined by the observed data, and the singular value decomposition is utilized to define a continuous time process that encodes the data´s intrinsic geometry. Two toy examples are given that demonstrate stochastic Markov matrices can preserve the underlying non-planar geometric geometry of the data and they provide an unsupervised tightening of natural cluster centers. Furthermore, these two properties of Markov chains are shown to improve an automated color image segmentation algorithm.