<title>Markov random fields on a SIMD machine for global region labelling</title>

Abstract

The Markov Random Field (MRF) formulation allows independence over small pixel neighborhoods suitable for SIMD implementation. The equivalence between the Gibbs distribution over global configurations and MRF allows describing the problem as maximizing a probability, or equivalently, minimizing an energy function (EF).The EF is a convenient device for integrating "votes" from disparate, pre-processed features-mean intensity, variance, moments, etc. Contributions from each feature are simply weighted and summed. The EF is flexible and can be easily modified to capture a priori beliefs about the distribution of the configuration space, and still remain theoretically sound. A unique formulation of the EF is given. Notably, a deterministic edge finder contributes to the EF. Weights are independently assigned to each feature's report (indicators).Simulated annealing is the theoretical mechanism which guarantees convergence in distribution to a global minimum. Because the number of iterations is an exponential function of time, we depart from theory and implement a fast, heuristic "cooling" schedule. A videotape of results on simulated FUR imagery demonstrates real-time update over the entire image. Actual convergence is still too slow for real-time use (O( 1 mm.)), but the quality of results compares favorably with other region labelling schemes.

Real-time DSP implementation for MRF-based video motion detection

“…For example in [6], an MRF-based global region labeling algorithm is implemented on a SIMD array of over 40 000 processing units. The keypoints in SIMD implementations are the distribution of data onto the processors and the communication between processors.…”