Take a look at the DESeq2 paper for an intuitive description of the fold change estimate. It's not a pseudocount based method, but a Bayesian method, which uses the information from the experiment to determine the best estimate when there is little information (eg. low counts, which can vary a lot).