Lip contour detection is regarded as an essential issue in many applications such as personal identification, facial expressions classification, and man-machine interaction. Moreover, semi-supervised learning is utilized to automatically exploit unlabeled data in addition to labeled data to improve the performance of certain machine learning approaches. In this paper, three contour preprocessing approaches for eliminate lip image noise, i.e., Average filtering, Bilateral filtering, and Edge preserving smoothing techniques are compared. Furthermore, a hybrid approach combing level set theory and semi-supervised Fisher transformation for lip contour detection is proposed. Experiment results show that the proposed semi-supervised strategy for lip contour detection is effective.