Automatic lip contour extraction using pixel-based segmentation and piece-wise polynomial fitting

This work presents automatic lip contour extraction using pixel-based segmentation and piece-wise polynomial fitting. In the first stage, the region of interest (ROI) i.e. mouth region is extracted by binary classification based on color ratio thresholding followed by centrally located large connected region detection. k-means clustering is applied to green plane to get upper lip area. The lower lip area is obtained by using binary k-means clustering to the weighted plane. The combined lip area is further processed to detect the centrally located big connected region. Robert filtering followed by similar neighbour traversing are employed to estimate the lip contour. A smoothed upper and lower contours are obtained by varying piece-wise polynomial fitting. Experimental results performed on standard GRID database show that the scheme performs well even under the influence of illumination and clothing effects.