The paper by Marroquin, Mitter, and Poggio formulates several problems in early vision as inverse problems and reviews standard regularization theory, discussing its limitations. They introduce new stochastic techniques based on Markov Random Field (MRF) models to solve these problems. The authors derive efficient algorithms and describe parallel implementations on digital parallel SIMD architectures and hybrid computers that mix digital and analog components. The paper focuses on the problem of surface reconstruction in early vision, where standard regularization methods struggle with discontinuities and fusion of information from multiple modules. The proposed Bayesian approach, using MRFs, effectively addresses these issues by representing a priori knowledge in a probability distribution and allowing for the computation of the posterior distribution. The algorithms are based on the Metropolis or Gibbs Sampler schemes to simulate the equilibrium behavior of the coupled MRF. The paper also discusses the estimation of parameters in noisy observations and the reconstruction of piecewise constant and continuous functions, demonstrating the effectiveness of the proposed methods through various examples and applications in computational vision.The paper by Marroquin, Mitter, and Poggio formulates several problems in early vision as inverse problems and reviews standard regularization theory, discussing its limitations. They introduce new stochastic techniques based on Markov Random Field (MRF) models to solve these problems. The authors derive efficient algorithms and describe parallel implementations on digital parallel SIMD architectures and hybrid computers that mix digital and analog components. The paper focuses on the problem of surface reconstruction in early vision, where standard regularization methods struggle with discontinuities and fusion of information from multiple modules. The proposed Bayesian approach, using MRFs, effectively addresses these issues by representing a priori knowledge in a probability distribution and allowing for the computation of the posterior distribution. The algorithms are based on the Metropolis or Gibbs Sampler schemes to simulate the equilibrium behavior of the coupled MRF. The paper also discusses the estimation of parameters in noisy observations and the reconstruction of piecewise constant and continuous functions, demonstrating the effectiveness of the proposed methods through various examples and applications in computational vision.