A prior parameter estimation method based on local center-encoding (LCE) is proposed for a Markov random field (MRF) model, i.e. the Ising case, in the task of image segmentation. The LCE algorithm makes efficient use of the local information in the image, avoiding the exclusion of certain blocks as in the least square (LSQR) algorithm. In addition, LCE doesn't require complex matrix computations, therefore reduces the computational cost. As a general algorithm, LCE can be used to estimate the prior parameters in anisotropic label fields. Experimental results on label fields and image segmentations demonstrate the efficiency and generality of the LCE algorithm.