Objective Correction for Policy Improvement under Entropy Regularization