The reading group paper will be:
Forbes, F., Peyrard, N. Hidden markov random field model selection criteria based on mean field-like approximations. PAMI 9/2003. pg. 1089- 1101. Paper here.
The problem setting is using MRFs to solve “low-level” vision problems, i.e. estimating a property at each patch/pixel in an image based on image data and the properties of surrounding patch/pixels. The specific problem the paper addresses is how to select statistical models for the observation and compatibility functions in MRFs. In particular, the point is that standard model selection criteria like BIC scores are difficult to compute directly for MRFs, but you can approximate these criteria.