General region merging based on first order Markov information theory statistical measures

A family of statistical region merging approaches based on a general region model is presented. Each region is modeled as an arbitrary first order finite-state Markov process, characterized by its empirical probability transition matrix. Under this premise, the merging problem is formulated from a statistical point of view, leading to two different merging criteria based on information theory statistical measures: the Kullback-Leibler divergence rate and the Bhattacharyya coefficient. In both cases, a size-independent extension of the previous methods, combined with a modified merging order, is also proposed. Finally, all methods are objectively evaluated and compared with other state-of-the-art region merging techniques.