Bayesian resolution-enhancement framework for transform-coded video