Abstract: In order to effectively exploit the lip feature of human speech, a lip segmentation method based on Markov random field (MRF) and model calibration is proposed. In this paper, we conduct the color space transformation from RGB to LUX color space for the lip region image, and we make use of the logarithmic chroma U to determine the initial contour. A mask with fixed radius is selected along the contour to define the local region, the Markov random field is used to segment the lips, and the Kullback-Leiller (KL) distance based model calibration method is used to coordinate the segmentation results between the local regions. Experiments show that the method can separate the lips in the skin with high accuracy and robustness and is of high practical value.