Most automatic scar segmentation methods for cardiac DE-CMR images rely on an existing myocardial segmentation (from CINE-CMR) that is registered to the DE-CMR volume, step where alignment errors are usually introduced. We present a variational method that, with the same inputs, identifies the healthy and scarred tissue and selectively corrects the endocardial and epicardial contours. For this, we tailor an existing multiphase segmentation method to provide different regularization costs for each region, and model the data fidelity energy term with a Bayesian approach that unifies the prior tissue probabilities and the myocardial labels. Experimental results show better overlapping for the ground truth and segmented myocardium, and the segmented scar compares favorably with respect to state of the art methods.