Face recognition is an important technique for Natural User Interface(NUI) and Human Robot Interaction(HRI) and many of the current state-of-the-art face recognition techniques are based on the local features which are extracted from a face alignment method like Constrained Local Model(CLM). But, in a real world environment, face alignment methods often fail to correctly localize the features because of extreme variations in pose and illumination. In this paper, we propose a learningbased misalinment detection and correction method. From the experiment, it is shown that the accuracy of the existing face alignment methods can be improved using the proposed method which re-aligns a misaligned result with a corrected parameter.